CN109408724A - Multimedia resource estimates the determination method, apparatus and server of clicking rate - Google Patents
Multimedia resource estimates the determination method, apparatus and server of clicking rate Download PDFInfo
- Publication number
- CN109408724A CN109408724A CN201811314076.6A CN201811314076A CN109408724A CN 109408724 A CN109408724 A CN 109408724A CN 201811314076 A CN201811314076 A CN 201811314076A CN 109408724 A CN109408724 A CN 109408724A
- Authority
- CN
- China
- Prior art keywords
- clicking rate
- multimedia resource
- information
- multimedia
- user
- 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
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
- G06Q30/0202—Market predictions or forecasting for commercial activities
-
- 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
- G06N3/045—Combinations of networks
-
- 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)
- Business, Economics & Management (AREA)
- Strategic Management (AREA)
- Data Mining & Analysis (AREA)
- Finance (AREA)
- Development Economics (AREA)
- Accounting & Taxation (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Entrepreneurship & Innovation (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Marketing (AREA)
- Game Theory and Decision Science (AREA)
- Economics (AREA)
- General Business, Economics & Management (AREA)
- Information Transfer Between Computers (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- User Interface Of Digital Computer (AREA)
Abstract
The disclosure is directed to determination method, apparatus and server that a kind of multimedia resource estimates clicking rate, belong to information recommendation field.The described method includes: obtaining the user behavior information of active user；The multimedia attribute information of the first multimedia resource is obtained, first multimedia resource is the multimedia resource to be recommended to the active user；Call clicking rate prediction model, the clicking rate prediction model includes that embeding layer and clicking rate estimate network, the embeding layer includes the corresponding weight matrix of at least one information type, the clicking rate estimates insertion vector of the network for exporting the embeding layer as input, and export multimedia resource estimates clicking rate；By clicking rate prediction model described in the user behavior information and the multimedia attribute information input, exports the user and clicking rate is estimated to first multimedia resource.Using the disclosure, the accuracy for estimating clicking rate can be improved.
Description
Technical field
This disclosure relates to which information recommendation field more particularly to a kind of multimedia resource estimate the determination method of clicking rate, dress
It sets and server.
Background technique
In information recommendation system, the relevant technologies can to multimedia resource clicking rate (Click Through Rate,
CTR it) is estimated, and the higher multimedia resource of clicking rate will be estimated and show user, click multimedia to improve user
The probability of resource improves the accuracy rate of information recommendation.
When being estimated to the clicking rate of multimedia resource, commonly uses be based on deep neural network (Deep in the related technology
Neural Network, DNN) clicking rate prediction model, calculate clicking rate is estimated.The input of clicking rate prediction model
Data can be divided into different fields, for example, the field in the field of multimedia resource and user.Clicking rate as shown in Figure 1 is pre-
Estimate model schematic, input data can be converted to the insertion vector of low-dimensional by insertion (embedding) layer, wherein different
Field can correspond to weight matrix different in embeding layer, the various features information in a field can be by corresponding
Weight matrix determines insertion vector.There are one-to-one relationships for characteristic information and insertion vector, therefore, the corresponding weight in field
Matrix is alternatively referred to as the embedding mapping table in the field.Finally, in the related technology can be by insertion vector input depth nerve
Network, export multimedia resource estimates clicking rate.
But in one field, the number of certain characteristic informations may be less, for example, when the feature in user field
When information includes the picture identification that user clicked, if the number that user clicks picture is less, collected picture identification
Number it is also less.When being trained to clicking rate prediction model, small numbers of characteristic information will lead to corresponding weight
The study of matrix is not enough, namely the representativeness of obtained insertion vector is weaker, so as to cause in the point to multimedia resource
When the rate of hitting is estimated, the obtained accuracy for estimating clicking rate is lower.
Summary of the invention
The disclosure provides the determination method, apparatus and server that a kind of multimedia resource estimates clicking rate, can solve pre-
Estimate the lower problem of the accuracy of clicking rate.
According to the first aspect of the embodiments of the present disclosure, a kind of determination method that multimedia resource estimates clicking rate is provided, packet
It includes:
Obtain the user behavior information of user；
The multimedia attribute information of the first multimedia resource is obtained, first multimedia resource is to be recommended to the use
The multimedia resource at family；
Clicking rate prediction model is called, the clicking rate prediction model includes that embeding layer and clicking rate estimate network, described
Embeding layer includes the corresponding weight matrix of at least one information type, and the clicking rate estimates network for the embeding layer is defeated
For insertion vector out as input, export multimedia resource estimates clicking rate；
Clicking rate described in the user behavior information and the multimedia attribute information input is estimated into network, described in output
User estimates clicking rate to first multimedia resource.
Optionally, described that clicking rate described in the user behavior information and the multimedia attribute information input is estimated into mould
Type exports the user and estimates clicking rate to first multimedia resource, comprising:
For every kind of information type, the information will be belonged in the user behavior information and the multimedia attribute information
The information of type inputs the corresponding weight matrix of information type described in the embeding layer, exports at least one insertion vector；
At least one insertion vector that the embeding layer is exported inputs the clicking rate prediction model, exports the user
Clicking rate is estimated to first multimedia resource.
Optionally, the training method of the clicking rate prediction model includes:
Obtain the initial model of the clicking rate prediction model；
Obtain at least one training sample, the training sample include the second multimedia resource multimedia attribute information,
User behavior information and sample of users when sample of users browses second multimedia resource is to more than second matchmaker
The click condition of body resource, the click condition include having clicked or not clicked on；
The initial model is trained based at least one described training sample, the clicking rate is obtained and estimates mould
Type.
Optionally, the initial model includes initial embeding layer and initial clicking rate estimates network, the initial insertion
Layer includes the corresponding initial weight matrix of at least one information type；
It is described that the initial model is trained based at least one described training sample, it obtains the clicking rate and estimates
Model, comprising:
Initial weight matrix corresponding for every kind of information type, based on the training sample comprising the information type to institute
It states initial weight matrix and carries out parameter adjustment, the corresponding weight matrix of the information type after being trained；
Network is estimated to the initial clicking rate based at least one described training sample and carries out parameter adjustment, is trained
Clicking rate afterwards estimates network；
It is estimated based on the corresponding weight matrix of at least one information type after training and the clicking rate after the training
Network obtains the clicking rate prediction model.
Optionally, in the parameter tuning process to the corresponding initial weight matrix of every kind of information type, when comprising
When the number of the training sample of first information type is less than the number of the training sample comprising the second information type, first letter
It ceases the corresponding learning rate of type and is greater than the corresponding learning rate of second information type.
Optionally, the information type includes works mark, author's mark and/or style identification.
Optionally, the user behavior information includes clicking historical information, paying close attention to information and/or like information, the point
The multimedia attribute information of multimedia resource of the historical information for indicating user's click is hit, the concern information is used for indicating
The multimedia attribute information of the multimedia resource of family concern, it is described to like information for indicating the favorite multimedia resource of user
Multimedia attribute information.
According to the second aspect of an embodiment of the present disclosure, the determining device that a kind of multimedia resource estimates clicking rate is provided, packet
It includes:
Acquiring unit is configured as obtaining the user behavior information of user；
The acquiring unit is additionally configured to obtain the multimedia attribute information of the first multimedia resource, more than described first
Media resource is the multimedia resource to be recommended to the user；
Call unit is configured as calling clicking rate prediction model, and the clicking rate prediction model includes embeding layer and point
The rate of hitting estimates network, and the embeding layer includes the corresponding weight matrix of at least one information type, and the clicking rate estimates network
For insertion vector for exporting the embeding layer as input, export multimedia resource estimates clicking rate；
Determination unit is configured as clicking rate described in the user behavior information and the multimedia attribute information input
Prediction model exports the user and estimates clicking rate to first multimedia resource.
Optionally, the determination unit, is configured as:
For every kind of information type, the information will be belonged in the user behavior information and the multimedia attribute information
The information of type inputs the corresponding weight matrix of information type described in the embeding layer, exports at least one insertion vector；
At least one insertion vector that the embeding layer exports is inputted into the clicking rate and estimates network, exports the user
Clicking rate is estimated to first multimedia resource.
Optionally, described device further includes training unit, and the training unit is configured as:
Obtain the initial model of the clicking rate prediction model；
Obtain at least one training sample, the training sample include the second multimedia resource multimedia attribute information,
User behavior information and sample of users when sample of users browses second multimedia resource is to more than second matchmaker
The click condition of body resource, the click condition include having clicked or not clicked on；
The initial model is trained based at least one described training sample, the clicking rate is obtained and estimates mould
Type.
Optionally, the initial model includes initial embeding layer and initial clicking rate estimates network, the initial insertion
Layer includes the corresponding initial weight matrix of at least one information type；
The training unit is configured as:
Initial weight matrix corresponding for every kind of information type, based on the training sample comprising the information type to institute
It states initial weight matrix and carries out parameter adjustment, the corresponding weight matrix of the information type after being trained；
Network is estimated to the initial clicking rate based at least one described training sample and carries out parameter adjustment, is trained
Clicking rate afterwards estimates network；
It is estimated based on the corresponding weight matrix of at least one information type after training and the clicking rate after the training
Network obtains the clicking rate prediction model.
Optionally, in the parameter tuning process to the corresponding initial weight matrix of every kind of information type, when comprising
When the number of the training sample of first information type is less than the number of the training sample comprising the second information type, first letter
It ceases the corresponding learning rate of type and is greater than the corresponding learning rate of second information type.
Optionally, the information type includes works mark, author's mark and/or style identification.
Optionally, the user behavior information includes clicking historical information, paying close attention to information and/or like information, the point
The multimedia attribute information of multimedia resource of the historical information for indicating user's click is hit, the concern information is used for indicating
The multimedia attribute information of the multimedia resource of family concern, it is described to like information for indicating the favorite multimedia resource of user
Multimedia attribute information.
According to the third aspect of an embodiment of the present disclosure, a kind of server is provided, comprising:
Processor；
Memory for storage processor executable instruction；
Wherein, the processor is configured to:
Obtain the user behavior information of user；
The multimedia attribute information of the first multimedia resource is obtained, first multimedia resource is to be recommended to the use
The multimedia resource at family；
Clicking rate prediction model is called, the clicking rate prediction model includes that embeding layer and clicking rate estimate network, described
Embeding layer includes the corresponding weight matrix of at least one information type, and the clicking rate estimates network for the embeding layer is defeated
For insertion vector out as input, export multimedia resource estimates clicking rate；
By clicking rate prediction model described in the user behavior information and the multimedia attribute information input, described in output
User estimates clicking rate to first multimedia resource.
According to a fourth aspect of embodiments of the present disclosure, a kind of non-transitorycomputer readable storage medium is provided, when described
When instruction in storage medium is executed by the processor of server, enables the server to execute a kind of multimedia resource and estimate a little
Hit the determination method of rate, which comprises
Obtain the user behavior information of user；
The multimedia attribute information of the first multimedia resource is obtained, first multimedia resource is to be recommended to the use
The multimedia resource at family；
Clicking rate prediction model is called, the clicking rate prediction model includes that embeding layer and clicking rate estimate network, described
Embeding layer includes the corresponding weight matrix of at least one information type, and the clicking rate estimates network for the embeding layer is defeated
For insertion vector out as input, export multimedia resource estimates clicking rate；
By clicking rate prediction model described in the user behavior information and the multimedia attribute information input, described in output
User estimates clicking rate to first multimedia resource.
According to a fifth aspect of the embodiments of the present disclosure, a kind of application program/computer program product is provided, when using journey
Sequence/computer program product server at runtime so that server executes a kind of multimedia resource and estimates clicking rate really
Determine method, which comprises
Obtain the user behavior information of active user；
The multimedia attribute information of the first multimedia resource is obtained, first multimedia resource is worked as to be to be recommended to described
The multimedia resource of preceding user；
Clicking rate prediction model is called, the clicking rate prediction model includes that embeding layer and clicking rate estimate network, described
Embeding layer includes the corresponding weight matrix of at least one information type, and the clicking rate estimates network for the embeding layer is defeated
For insertion vector out as input, export multimedia resource estimates clicking rate；
By clicking rate prediction model described in the user behavior information and the multimedia attribute information input, described in output
User estimates clicking rate to first multimedia resource.
The technical scheme provided by this disclosed embodiment can include the following benefits:
In user behavior information and multimedia attribute information, the information of identical information type can pass through phase in embeding layer
Same weight matrix determines insertion vector, and the representativeness of insertion vector can be improved.To in the method pair based on the present embodiment
When the clicking rate of multimedia resource is estimated, the accuracy for estimating clicking rate is improved.
It should be understood that above general description and following detailed description be only it is exemplary and explanatory, not
The disclosure can be limited.
Detailed description of the invention
The drawings herein are incorporated into the specification and forms part of this specification, and shows the implementation for meeting the disclosure
Example, and together with specification for explaining the principles of this disclosure.
Fig. 1 is a kind of clicking rate prediction model schematic diagram shown according to an exemplary embodiment.
Fig. 2 is a kind of implementation environment figure shown according to an exemplary embodiment.
Fig. 3 is the determination method flow that a kind of multimedia resource shown according to an exemplary embodiment estimates clicking rate
Figure.
Fig. 4 is the determination method flow that a kind of multimedia resource shown according to an exemplary embodiment estimates clicking rate
Figure.
Fig. 5 is a kind of Application Program Interface schematic diagram shown according to an exemplary embodiment.
Fig. 6 is a kind of clicking rate prediction model schematic diagram shown according to an exemplary embodiment.
Fig. 7 is a kind of clicking rate prediction model schematic diagram shown according to an exemplary embodiment.
Fig. 8 is a kind of training method flow chart of clicking rate prediction model shown according to an exemplary embodiment.
Fig. 9 is a kind of training method flow chart of clicking rate prediction model shown according to an exemplary embodiment.
Figure 10 is the determination device block diagram that a kind of multimedia resource shown according to an exemplary embodiment estimates clicking rate.
Figure 11 is shown according to an exemplary embodiment a kind of for determining that multimedia resource estimates the device of clicking rate
Block diagram.
Specific embodiment
Example embodiments are described in detail here, and the example is illustrated in the accompanying drawings.Following description is related to
When attached drawing, unless otherwise indicated, the same numbers in different drawings indicate the same or similar elements.Following exemplary embodiment
Described in embodiment do not represent all implementations consistent with this disclosure.On the contrary, they be only with it is such as appended
The example of the consistent device and method of some aspects be described in detail in claims, the disclosure.
Present embodiments provide the implementation environment figure that a kind of multimedia resource estimates the determination method of clicking rate, the implementation ring
Border figure is as shown in Figure 2.The implementation environment may include multiple terminals 201, for providing the service of service for multiple terminal 201
Device 202.Multiple terminals 201 are connected by wireless or cable network and server 202, and multiple terminal 201 can be can
Access computer equipment or the intelligent terminal etc. of server 202.It can be equipped in terminal 201 for recommending multimedia resource
The application program of (such as picture, short-sighted frequency), user can log in above-mentioned application program.Server 202 can be above-mentioned application
Program provides background service, and records the user behavior information of each user.There can also be at least one number in server 202
According to library, to store user's row of clicking rate prediction model, multimedia resource and corresponding multimedia attribute information, each user
For information etc..
A kind of determination method that multimedia resource estimates clicking rate is present embodiments provided, this method can be by server reality
Existing, multimedia resource as shown in Figure 3 estimates the determination method flow diagram of clicking rate, the process flow of this method may include as
Under step:
In step S301, server obtains the user behavior information of user.
In step s 302, server obtains the multimedia attribute information of the first multimedia resource.
Wherein, the first multimedia resource is the multimedia resource to be recommended to user.
In step S303, server calls clicking rate prediction model.
Wherein, clicking rate prediction model includes embeding layer and clicking rate estimates network, and embeding layer includes at least one information
The corresponding weight matrix of type, clicking rate estimate insertion vector of the network for exporting embeding layer as input, export more matchmakers
Body resource estimates clicking rate.
In step s 304, server is by user behavior information and multimedia attribute information input clicking rate prediction model,
Output user estimates clicking rate to the first multimedia resource.
Optionally, described that user behavior information and multimedia attribute information input clicking rate are estimated into network, export user
Clicking rate of estimating to the first multimedia resource includes:
For every kind of information type, the letter of information type will be belonged in user behavior information and the multimedia attribute information
Breath inputs the corresponding weight matrix of information type in embeding layer, exports at least one insertion vector；
At least one insertion vector of embeding layer output is inputted into clicking rate prediction model, exports user to the first multimedia
Resource estimates clicking rate.
Optionally, the training method of clicking rate prediction model includes:
Obtain the initial model of clicking rate prediction model；
At least one training sample is obtained, training sample includes the multimedia attribute information of the second multimedia resource, sample
User behavior information and sample of users when user browses the second multimedia resource to the click condition of the second multimedia resource,
Click condition includes having clicked or not clicked on；
Initial model is trained based at least one training sample, obtains clicking rate prediction model.
Optionally, initial model includes initial embeding layer and initial clicking rate estimates network, and initial embeding layer includes extremely
The corresponding initial weight matrix of a kind of information type less；
It is described that initial model is trained based at least one training sample, obtain clicking rate prediction model, comprising:
Initial weight matrix corresponding for every kind of information type, based on the training sample comprising information type to initial power
Weight matrix carries out parameter adjustment, the corresponding weight matrix of information type after being trained；
Network is estimated to initial clicking rate based at least one training sample and carries out parameter adjustment, the click after being trained
Rate estimates network；
Network is estimated based on the corresponding weight matrix of at least one information type after training and the clicking rate after training,
Obtain clicking rate prediction model.
Optionally, in the parameter tuning process to the corresponding initial weight matrix of every kind of information type, when including first
When the number of the training sample of information type is less than the number of the training sample comprising the second information type, first information type pair
The learning rate answered is greater than the corresponding learning rate of second information type.
Optionally, information type includes works mark, author's mark and/or style identification.
Optionally, user behavior information includes clicking historical information, paying close attention to information and/or like information, clicks history letter
Breath includes that user clicks the information recorded when any multimedia resource, and concern information includes that user records when clicking concern option
Information likes that information includes that user clicks the information for liking recording when option.
The present embodiment will be situated between in conjunction with specific embodiment, the determination method for estimating clicking rate to multimedia resource
It continues.This method can be realized that multimedia resource as shown in Figure 4 estimates the determination method flow diagram of clicking rate by server, should
The process flow of method may include following step:
In step S401, server obtains the user behavior information of user.
When user opens application program, terminal can send the acquisition request of multimedia resource to server；Alternatively, user
When by application program searching multimedia resource, terminal can also send the acquisition request of multimedia resource to server.Service
When device receives the acquisition request of the multimedia resource of terminal transmission, the processing logic for recommending multimedia resource can be triggered, with
It will pass through multimedia resource provided in this embodiment and estimate the determination method of clicking rate and handled.The present embodiment recommends triggering
The concrete mode of the processing logic of multimedia resource is not construed as limiting.
At this point, server can be obtained from the user behavior information of each user of storage according to the identification information of user
Take the user behavior information of the user.In a kind of possible embodiment, user behavior information may include clicking history letter
It ceases, pay close attention to information and/or like information, click the multimedia for the multimedia resource that historical information can be used to indicate that user clicks
Attribute information, concern information can be used to indicate that the multimedia attribute information of the multimedia resource of user's concern, likes information can
For indicating the multimedia attribute information of the favorite multimedia resource of user.
Optionally, for the multimedia attribute information of the multimedia resource of server record, information type may include making
Product mark, author's mark and/or style identification.
The process of server record above- mentioned information is introduced below:
Application Program Interface schematic diagram as shown in Figure 5, terminal can show the multimedia resource that application program provides, and
It may include paying close attention to option and liking option and in the displaying interface of multimedia resource.
When checking the multimedia resource when the user clicks, server can receive the request for loading the multimedia resource,
And then corresponding multimedia attribute information can be added in the click historical information of the user and be stored.For example, working as
When first short-sighted frequency is watched in user's click, server can be identified the works of the short-sighted frequency, author identifies and/or style identification
It is added in the click historical information of the user.
If user needs to subscribe to the multimedia resource, the concern option shown in interface can be clicked.In turn, it services
Device can receive the addition request of the concern to the multimedia resource, and corresponding multimedia attribute information is added to the user's
It is stored in concern information.Certainly, concern option is directed to the concern of multimedia resource in addition to can be, and is also possible to needle
The concern of author or style for multimedia resource, the present embodiment are not construed as limiting this.For example, if concern option is to be directed to
In the concern of the author of multimedia resource, then the author of the multimedia resource can be identified the concern for being added to user by server
In information, after this, when author's more new works, user can receive corresponding update notification, reach the effect of subscription
Fruit.
If user likes the multimedia resource, the option of liking shown in interface, server addition pair can be clicked
Similarly, details are not described herein again for the processing for liking information answered and concern information.
Certainly, above-mentioned record clicks historical information, pays close attention to information and like that the scheme of information is optinal plan, that is,
The user behavior information of server record can be and click historical information, one of pays close attention to information and like information or a variety of.
Information type wherein included is equally also possible to one of works mark, author's mark and style identification or a variety of, may be used also
To include other information type, the present embodiment is not construed as limiting specific information type.
In step S402, server obtains the multimedia attribute information of the first multimedia resource.
Wherein, the first multimedia resource is the multimedia resource to be recommended to user.Multimedia attribute information can be
Actively addition obtains or author of the server according to multimedia resource, content author when author uploads corresponding multimedia resource
Equal Automatic generation of information, the present embodiment are not construed as limiting the generating mode of multimedia attribute information.
First multimedia resource may include multiple multimedia resources, for example, it may be current popular multimedia resource,
Or it can be the search result obtained to the search of user.Since the multimedia resource that user can check is limited, server
It can determine which the first multimedia resource shown, and the first multimedia resource of displaying can also be ranked up, so as to
Achieve the purpose that preferentially to show the multimedia resource for being more in line with user demand.The pre- of multimedia resource is utilized in the present embodiment
Clicking rate is estimated to realize above-mentioned purpose, it is bigger to estimate a possibility that clicking rate is higher, and user clicks the multimedia resource, namely be somebody's turn to do
Multimedia resource more meets the demand of user.
Determining when estimating clicking rate of the first multimedia resource, for first multimedia resource, server can be with
It is obtained from the multimedia resource of storage and corresponding multimedia attribute information according to the identification information of first multimedia resource
Take the multimedia attribute information of first multimedia resource.
In step S403, server calls clicking rate prediction model.
Clicking rate prediction model schematic diagram as shown in FIG. 6, clicking rate prediction model provided in this embodiment may include
Embeding layer and clicking rate estimate network, and embeding layer may include the corresponding weight matrix of at least one information type, and clicking rate is pre-
Estimating network can be used for using the insertion vector of embeding layer output as input, and export multimedia resource estimates clicking rate.It clicks
Estimating network can be deep neural network, be also possible to convolutional neural networks etc., and technical staff can be according to demand to click
Rate is estimated network and is designed, and the present embodiment is not construed as limiting specific network structure.
Certainly, multiple clicking rate prediction models also can store in server, server can call the default item of satisfaction
The clicking rate prediction model of part, for example, can be adjusted when being also stored with the accuracy rate of each clicking rate prediction model in server
Any clicking rate prediction model for being greater than preset threshold with accuracy rate is met.The present embodiment is to the specific side for calling clicking rate model
Formula is not construed as limiting.
Clicking rate prediction model can be periodically trained, and can store the newest point that training obtains in server
Hit rate prediction model.When server it needs to be determined that when estimating clicking rate of the first multimedia resource, can call newest click
Rate prediction model.
In step s 404, for every kind of information type, server will be in user behavior information and multimedia attribute information
Belong to the information of the information type, input the corresponding weight matrix of the information type in embeding layer, export at least one be embedded in
Amount.
Server can be got the user behavior information got in above-mentioned steps S401 and in step S402
The multimedia attribute information of first multimedia resource inputs the embeding layer of clicking rate prediction model.
In embeding layer, input data can be encoded, to adapt to the calculating of neural network.In general, compiling
The vector dimension obtained after code is larger, therefore can be carried out at dimensionality reduction by the weight matrix in embeding layer to the vector after coding
Reason, to reduce the IO in treatment process (Input-Output, input/output) expense.For example, to a Xiang Te of input data
Reference breath obtains vector (0,0,0,1,0,0,0,0,0) after carrying out one-hot coding, is converted to insertion vector by weight matrix
(0.145,0.152)。
Every kind of information type has corresponding weight matrix, in user behavior information and multimedia attribute information, service
The corresponding vector of characteristic information of the available identical information type of device, the corresponding vector input of same information type is corresponding
Weight matrix, output insertion vector.
It is had been described above in above-mentioned steps S401, in a kind of possible embodiment, information type may include works
Mark, author's mark and/or style identification.Illustratively, clicking rate prediction model schematic diagram as shown in Figure 7, user behavior
Information is to click historical information, wherein each characteristic information is the works mark that user clicked, the spy of multimedia attribute information
Reference breath is that the works of the multimedia resource identify.Embedding mapping table shown in Fig. 7 is that works identify corresponding weight square
Battle array, server can identify the works in user behavior information and the works of multimedia attribute information identify, by being somebody's turn to do
Embedding mapping table determines insertion vector.That is, the weight matrix phase in embeding layer used in the information of works mark
Together, it will not be distinguished with the field in the field of multimedia resource or user.
In step S405, at least one insertion vector input clicking rate that embeding layer exports is estimated network by server,
Output user estimates clicking rate to the first multimedia resource.
After server determines each insertion vector of input data, insertion vector input clicking rate can be estimated into net
Network estimates each network node in network by clicking rate and carries out data processing, and the first multimedia resource of output is estimated a little
Hit rate.
For each first multimedia resource, S402-S405 determination clicking rate can be estimated through the above steps.Multiple
One multimedia resource can be the determination by way of parallel processing and estimate clicking rate, and the present embodiment provides multiple first multimedias
Source determines that the sequence for estimating clicking rate is not construed as limiting.
Server determine under each first multimedia resource estimate clicking rate after, can according to estimate clicking rate from greatly to
Small sequence is ranked up each first multimedia resource, and can be by each first multimedia resource and corresponding row
Sequence is sent to terminal.In turn, terminal can be shown the first multimedia resource received according to the sequence.Optionally,
When the number of the multimedia resource of displaying is preset number, server can will sort forward preset number more than first matchmaker
Body resource and corresponding sequence are sent to terminal, and terminal can be according to the sequence to the first multimedia resource of the preset number
It is shown.
Certainly, the first multimedia resource that server is sent to terminal can be the corresponding breviary shape of the first multimedia resource
Formula, to reduce the consumption of Internet resources, for example, what server was sent to terminal can when the first multimedia resource is picture
To be the thumbnail of picture；When the first multimedia resource is short-sighted frequency, server can be preview graph to what terminal was sent.This
The concrete form for the first multimedia resource that embodiment is sent is not construed as limiting.
After the first multimedia resource of terminal display, what active user preferentially saw, which can be, estimates clicking rate higher first
Multimedia resource.Therefore, the multimedia resource provided through this embodiment estimates the determination method of clicking rate, and user can be improved
To the clicking rate of the multimedia resource of recommendation.When the above method is applied in the application, the use of application program can be improved
Family retention ratio.
In the present embodiment, in user behavior information and multimedia attribute information, the information of identical information type can lead to
It crosses identical weight matrix in embeding layer and determines insertion vector, the representativeness of insertion vector can be improved.To be based on this reality
When the method for applying example estimates the clicking rate of multimedia resource, the accuracy rate for estimating clicking rate is improved.
The process for determining using clicking rate prediction model and estimating clicking rate is described in above-described embodiment, is using clicking rate
Before prediction model, which can be trained.Present embodiments provide a kind of clicking rate prediction model
Training method, this method can realize by server.The training method flow chart of clicking rate prediction model as shown in Figure 8,
The process flow of this method may include following step:
In step S801, server obtains the initial model of clicking rate prediction model.
Wherein, corresponding with the clicking rate prediction model introduced in above-described embodiment, initial model may include initial
Embeding layer and initial clicking rate estimate network, and initial embeding layer includes the corresponding initial weight square of at least one information type
Battle array.
It can store the initial model of clicking rate prediction model in server.The initial model can be technical staff and set
Meter estimates the machine learning model of clicking rate for determining, using user behavior information and multimedia attribute information as defeated
Enter, prediction user is to the clicking rate of multimedia resource, and output estimation clicking rate.But since the model parameter in initial model is equal
Clicking rate accuracy for preset initial value, prediction is lower, needs to be trained initial model.
In step S802, server obtains at least one training sample.
Wherein, training sample may include the multimedia attribute information of the second multimedia resource, sample of users browsing second
To the click condition of the second multimedia resource, click condition can be with for user behavior information and sample of users when multimedia resource
Including having clicked or not clicked on.That is, the second multimedia resource can refer to that history shows the multimedia resource of sample of users.
The process of the information in server record training sample is introduced below:
When server sends the second multimedia resource for display to terminal, second multimedia money can recorde
The multimedia attribute information in source.Certainly, server also can recorde the identification information of second multimedia resource, the identification information
It can be used for obtaining the multimedia attribute information of the second multimedia resource.
At this point, server can also obtain the user behavior information of user, and can be by the user behavior information and clothes
Above- mentioned information (the multimedia attribute information of such as the second multimedia resource or the mark of the second multimedia resource of business device record
Information) it corresponding is recorded.
When checking any second multimedia resource when the user clicks, server, which can receive, loads second multimedia money
The click condition of second multimedia resource can be recorded as having clicked by the request in source in turn, it is corresponding with above- mentioned information into
Row record, for example, by the multimedia attribute information, user behavior information and click condition of the second multimedia resource it is corresponding into
Row record.
When terminal closes displaying interface, it can be sent to server and show closing notice, server receive the displaying and close
When closing notice, the second multimedia resource not clicked on can be obtained in the second multimedia resource of transmission, and by corresponding point
Situation is hit to be recorded as not clicking on.Alternatively, server can also after sending the second multimedia resource for display to terminal,
When reaching preset duration, the second multimedia resource not clicked on is obtained.The present embodiment obtains more than second not clicked on to server
The concrete mode of media resource is not construed as limiting.
When server is trained clicking rate prediction model, the second multimedia for being recorded in the available above process
User behavior information and user when the multimedia attribute information of resource, user browse the second multimedia resource is to more than second matchmaker
The click condition of body resource, as training sample.Optionally, it is that the training sample clicked is made that server, which can will click on situation,
For positive sample, will click on situation is the training sample that does not click on as negative sample.
In step S803, server is based at least one training sample and is trained to initial model, obtains clicking rate
Prediction model.
For each training sample, server can will be at the beginning of multimedia attribute information therein and user behavior information input
Beginning model, and data processing is carried out based on the model parameter of network node each in initial model, the initial model is obtained to the
Two multimedia resources estimate clicking rate.Then, server can be according to user in training sample to the second multimedia resource
Click condition and it is corresponding estimate clicking rate, determine the gradient of each model parameter in initial model.Server can basis
The gradient of each model parameter is determined the correction value of each model parameter, and is joined based on correction value to each model parameter
Number adjustment namely error back propagation.
In a kind of possible embodiment, the training method flow chart of clicking rate prediction model as shown in Figure 9, step
S803 may include step S8031-S8033:
In step S8031, initial weight matrix corresponding for every kind of information type, server is based on including info class
The training sample of type carries out parameter adjustment to initial weight matrix, the corresponding weight matrix of information type after being trained.
Each model parameter in weight matrix can be adjusted in the training process.By the agency of in above-described embodiment
, the characteristic information of identical information type can determine insertion vector based on identical weight matrix, and then click is estimated in determination
Rate.Corresponding, in the training process, initial weight matrix corresponding for every kind of information type, server is available to be led to
Cross the initial weight matrix determine the training sample for estimating clicking rate and it is corresponding estimate clicking rate, and then can be according to pre-
Estimate clicking rate and the actual click condition of user, the gradient of each model parameter in the initial weight matrix is calculated, according to gradient
It determines corresponding correction value, each model parameter is adjusted.After training, the obtained corresponding power of each information type
Weight matrix can be used to determine corresponding information insertion vector.
Compared to the scheme that the field in field and user based on multimedia resource divides weight matrix, due to identical information
The information of type uses same weight matrix, can make full use of the training sample comprising the information type to weight matrix into
Row training, so that the weight matrix is adequately learnt.
In step S8032, server is based at least one training sample and estimates network progress parameter to initial clicking rate
Adjustment, the clicking rate after being trained estimate network.
In general, each training sample can estimate network by clicking rate and carry out data processing, therefore, server
Available each training sample and it is corresponding estimate clicking rate, to initial clicking rate estimate network carry out parameter adjustment, tool
Body processing is as described above, details are not described herein again.
In step S8033, server is based on the corresponding weight matrix of at least one information type and instruction after training
Clicking rate after white silk estimates network, obtains clicking rate prediction model.
When reaching the condition (value for such as reaching default frequency of training or loss function is less than target value) of training end,
Each weight matrix and clicking rate in the available currently embedded layer of server estimate network etc., constitute clicking rate and estimate
Model, and the clicking rate prediction model can be stored.It, can when server needs to predict multimedia resource
It is handled with obtaining stored clicking rate prediction model.
Certainly, after this, server can also be trained again the clicking rate prediction model of storage, training process
Similarly with the above process.Server is constantly updated clicking rate prediction model, and the standard of clicking rate prediction model can be improved
True property.
In a kind of possible embodiment, the above-mentioned parameter tune to the corresponding initial weight matrix of every kind of information type
During whole, when the number of the training sample comprising first information type is less than the number of the training sample comprising the second information type
When mesh, the corresponding learning rate of first information type is greater than the corresponding learning rate of the second information type.
In above-mentioned steps S8031, when determining the correction value of each model parameter in initial weight matrix according to gradient, clothes
Business device can be adjusted learning rate according to gradient, for example, the method for learning rate adjustment can be AdaGrad (Adaptive
Gradient, autoadapted learning rate) algorithm.
In general, if the training sample comprising information type is less, it is determined that gradient is smaller namely change of gradient
More gentle, server can increase the corresponding learning rate of the information type, namely the amplitude of correction value is increased, and enable weight
Matrix obtains more fully gradient updating.By the above method, it can make model parameter can also in the case where feature is sparse
Adequately to be learnt, the accuracy rate of model is improved.
Certainly, the method for above-mentioned regularized learning algorithm rate can also be applied in above-mentioned steps S8032, so that clicking rate estimates net
Model parameter in network can also be adjusted according to gradient preference, and the present embodiment estimates the parameter regulation means of network to clicking rate
It is not construed as limiting.
In a kind of possible embodiment, server may be such that AUC (Area to the training objective of initial model
Under the ROC Curve, ROC curve area under；ROC, Receiver Operating Characteristic, by
Examination person's operating characteristic) it maximizes.
If will click on situation is that the training sample clicked is known as the first training sample, it will click on situation and do not click on
Training sample is known as the second training sample, then AUC can refer to that the first training sample comes the probability before the second training sample.
Server by initial model to each training sample determine estimate clicking rate after, can according to estimate clicking rate from
It arrives small sequence greatly to be arranged, and then can be according to the number for coming the first training sample before all second training samples
Mesh and training sample sum, determine the numerical value of AUC.AUC is bigger, shows that the first more training samples comes all second
Before training sample namely the accuracy rate of clicking rate prediction model is higher.
Certainly, server is also based on other way and determines AUC, such as can estimate clicking rate based on training sample
After establishing ROC curve, the area below ROC curve is calculated using the method quadratured.The present embodiment is to the specific side for determining AUC
Formula is not construed as limiting.
Experiment shows that method provided in this embodiment enables to AUC to be obviously improved, namely through this embodiment
The clicking rate prediction model accuracy rate that method obtains improves.
In the present embodiment, weight matrix corresponding for an information type can be based on the instruction comprising the information type
Practice sample to be trained.Since the feature of same information type is believed in the field and the field of user that are not based on multimedia resource
Breath is divided, therefore can make full use of training sample, so that the weight matrix of embeding layer is adequately learnt, is improved embedding
The representativeness of incoming vector, and then improve the accuracy rate of clicking rate prediction model.
Figure 10 is the determination device block diagram that a kind of multimedia resource shown according to an exemplary embodiment estimates clicking rate.
Referring to Fig.1 0, which includes acquiring unit 1010, call unit 1020 and determination unit 1030.
The acquiring unit 1010 is configured as obtaining the user behavior information of user；
The acquiring unit 1010, be additionally configured to obtain the first multimedia resource multimedia attribute information, described first
Multimedia resource is the multimedia resource to be recommended to the user；
The call unit 1020 is configured as calling clicking rate prediction model, and the clicking rate prediction model includes insertion
Layer and clicking rate estimate network, and the embeding layer includes the corresponding weight matrix of at least one information type, and the clicking rate is pre-
Insertion vector of the network for exporting the embeding layer is estimated as input, and export multimedia resource estimates clicking rate；
The determination unit 1030, being configured as will be described in the user behavior information and the multimedia attribute information input
Clicking rate prediction model exports the user and estimates clicking rate to first multimedia resource.
Optionally, the determination unit 1030, is configured as:
For every kind of information type, the information will be belonged in the user behavior information and the multimedia attribute information
The information of type inputs the corresponding weight matrix of information type described in the embeding layer, exports at least one insertion vector；
At least one insertion vector that the embeding layer is exported inputs the clicking rate prediction model, exports the user
Clicking rate is estimated to first multimedia resource.
Optionally, the device further includes training unit, which is configured as:
Obtain the initial model of the clicking rate prediction model；
Obtain at least one training sample, the training sample include the second multimedia resource multimedia attribute information,
User behavior information and user when user browses second multimedia resource is to second multimedia resource
Click condition, the click condition include having clicked or not clicked on；
The initial model is trained based at least one described training sample, the clicking rate is obtained and estimates mould
Type.
Optionally, the initial model includes initial embeding layer and initial clicking rate estimates network, the initial insertion
Layer includes the corresponding initial weight matrix of at least one information type；
The training unit is configured as:
Initial weight matrix corresponding for every kind of information type, based on the training sample comprising the information type to institute
It states initial weight matrix and carries out parameter adjustment, the corresponding weight matrix of the information type after being trained；
Network is estimated to the initial clicking rate based at least one described training sample and carries out parameter adjustment, is trained
Clicking rate afterwards estimates network；
It is estimated based on the corresponding weight matrix of at least one information type after training and the clicking rate after the training
Network obtains the clicking rate prediction model.
Optionally, in the parameter tuning process to the corresponding initial weight matrix of every kind of information type, when comprising
When the number of the training sample of first information type is less than the number of the training sample comprising the second information type, first letter
It ceases the corresponding learning rate of type and is greater than the corresponding learning rate of second information type.
Optionally, the information type includes works mark, author's mark and/or style identification.
Optionally, the user behavior information includes clicking historical information, paying close attention to information and/or like information, the point
The multimedia attribute information of multimedia resource of the historical information for indicating user's click is hit, the concern information is used for indicating
The multimedia attribute information of the multimedia resource of family concern, it is described to like information for indicating the favorite multimedia resource of user
Multimedia attribute information.
About the device in above-described embodiment, wherein modules execute the concrete mode of operation in related this method
Embodiment in be described in detail, no detailed explanation will be given here.
In the present embodiment, in user behavior information and multimedia attribute information, the information of identical information type can lead to
It crosses identical weight matrix in embeding layer and determines insertion vector, the representativeness of insertion vector can be improved.To be based on this reality
When the method for applying example estimates the clicking rate of multimedia resource, the accuracy rate of clicking rate prediction model is improved.
Figure 11 is shown according to an exemplary embodiment a kind of for determining that multimedia resource estimates the device of clicking rate
1100 block diagram.For example, device 1100 may be provided as a server.Referring to Fig.1 1, device 1100 includes processing component
1122, it further comprise one or more processors, and the memory resource as representated by memory 1132, for storing
It can be by the instruction of the execution of processing component 1122, such as application program.The application program stored in memory 1132 may include
It is one or more each correspond to one group of instruction module.In addition, processing component 1122 is configured as executing instruction,
To execute the determination method that above-mentioned multimedia resource estimates clicking rate.
Device 1100 can also include that a power supply module 1126 be configured as the power management of executive device 1100, and one
Wired or wireless network interface 1150 is configured as device 1100 being connected to network and input and output (I/O) interface
1158.Device 1100 can be operated based on the operating system for being stored in memory 1132, such as Windows ServerTM, Mac
OS XTM, UnixTM, LinuxTM, FreeBSDTM or similar.
In the exemplary embodiment, a kind of non-transitorycomputer readable storage medium is additionally provided, for example including instruction
Memory, above-metioned instruction can by the processor in server execute to complete the determination that above-mentioned multimedia resource estimates clicking rate
Method.For example, the computer readable storage medium can be ROM, random access memory (RAM), CD-ROM, tape, soft
Disk and optical data storage devices etc..
In the exemplary embodiment, a kind of application program/computer program product is additionally provided, including one or more refers to
It enables, which can be executed by the processor of server, estimate clicking rate to complete above-mentioned multimedia resource
Determine method.
Those skilled in the art will readily occur to its of the disclosure after considering specification and practicing this disclosure
Its embodiment.This application is intended to cover any variations, uses, or adaptations of the disclosure, these modifications, purposes or
Person's adaptive change follows the general principles of this disclosure and including the undocumented common knowledge in the art of the disclosure
Or conventional techniques.The description and examples are only to be considered as illustrative, and the true scope and spirit of the disclosure are by following
Claim is pointed out.
It should be understood that the present disclosure is not limited to the precise structures that have been described above and shown in the drawings, and
And various modifications and changes may be made without departing from the scope thereof.The scope of the present disclosure is only limited by the accompanying claims.
Claims (10)
1. a kind of determination method that multimedia resource estimates clicking rate characterized by comprising
Obtain the user behavior information of user；
The multimedia attribute information of the first multimedia resource is obtained, first multimedia resource is to be recommended to the user's
Multimedia resource；
Clicking rate prediction model is called, the clicking rate prediction model includes embeding layer and clicking rate estimates network, the insertion
Layer includes the corresponding weight matrix of at least one information type, and the clicking rate estimates what network was used to export the embeding layer
Vector is embedded in as input, export multimedia resource estimates clicking rate；
By clicking rate prediction model described in the user behavior information and the multimedia attribute information input, the user is exported
Clicking rate is estimated to first multimedia resource.
2. the method according to claim 1, wherein described by the user behavior information and the multimedia category
Property information input described in clicking rate prediction model, export the user to the clicking rate of estimating of first multimedia resource, packet
It includes:
For every kind of information type, the information type will be belonged in the user behavior information and the multimedia attribute information
Information, input the corresponding weight matrix of information type described in the embeding layer, export at least one insertion vector；
At least one insertion vector that the embeding layer exports is inputted into the clicking rate and estimates network, exports the user to institute
That states the first multimedia resource estimates clicking rate.
3. the method according to claim 1, wherein the training method of the clicking rate prediction model includes:
Obtain the initial model of the clicking rate prediction model；
At least one training sample is obtained, the training sample includes the multimedia attribute information of the second multimedia resource, sample
User behavior information and the sample of users when user browses second multimedia resource provide second multimedia
The click condition in source, the click condition include having clicked or not clicked on；
The initial model is trained based at least one described training sample, obtains the clicking rate prediction model.
4. according to the method described in claim 3, it is characterized in that, the initial model includes initial embeding layer and initial point
The rate of hitting estimates network, and the initial embeding layer includes the corresponding initial weight matrix of at least one information type；
It is described that the initial model is trained based at least one described training sample, it obtains the clicking rate and estimates mould
Type, comprising:
Initial weight matrix corresponding for every kind of information type, based on the training sample comprising the information type to described first
Beginning weight matrix carries out parameter adjustment, the corresponding weight matrix of the information type after being trained；
Network is estimated to the initial clicking rate based at least one described training sample and carries out parameter adjustment, after being trained
Clicking rate estimates network；
Network is estimated based on the corresponding weight matrix of at least one information type after training and the clicking rate after the training,
Obtain the clicking rate prediction model.
5. according to the method described in claim 4, it is characterized in that, to the corresponding initial weight square of every kind of information type
In the parameter tuning process of battle array, when the number of the training sample comprising first information type is less than the instruction comprising the second information type
When practicing the number of sample, the corresponding learning rate of the first information type is greater than the corresponding learning rate of second information type.
6. -5 any method according to claim 1, which is characterized in that the information type includes works mark, author
Mark and/or style identification.
7. -5 any method according to claim 1, which is characterized in that the user behavior information includes clicking history letter
It ceases, pay close attention to information and/or like information, the multimedia for clicking the multimedia resource that historical information is used to indicate that user clicks
Attribute information, the concern information is used to indicate the multimedia attribute information of the multimedia resource of user's concern, described to like letter
Cease the multimedia attribute information for indicating the favorite multimedia resource of user.
8. the determining device that a kind of multimedia resource estimates clicking rate characterized by comprising
Acquiring unit is configured as obtaining the user behavior information of user；
The acquiring unit is additionally configured to obtain the multimedia attribute information of the first multimedia resource, first multimedia
Resource is the multimedia resource to be recommended to the user；
Call unit is configured as calling clicking rate prediction model, and the clicking rate prediction model includes embeding layer and clicking rate
Network is estimated, the embeding layer includes the corresponding weight matrix of at least one information type, and the clicking rate is estimated network and is used for
For the insertion vector that the embeding layer is exported as input, export multimedia resource estimates clicking rate；
Determination unit is configured as estimating the user behavior information and clicking rate described in the multimedia attribute information input
Model exports the user and estimates clicking rate to first multimedia resource.
9. a kind of server characterized by comprising
Processor；
Memory for storage processor executable instruction；
Wherein, the processor is configured to:
Obtain the user behavior information of user；
The multimedia attribute information of the first multimedia resource is obtained, first multimedia resource is to be recommended to the user's
Multimedia resource；
Clicking rate prediction model is called, the clicking rate prediction model includes embeding layer and clicking rate estimates network, the insertion
Layer includes the corresponding weight matrix of at least one information type, and the clicking rate estimates what network was used to export the embeding layer
Vector is embedded in as input, export multimedia resource estimates clicking rate；
By clicking rate prediction model described in the user behavior information and the multimedia attribute information input, the user is exported
Clicking rate is estimated to first multimedia resource.
10. a kind of non-transitorycomputer readable storage medium, which is characterized in that when the instruction in the storage medium is by servicing
When the processor of device executes, enable the server to execute a kind of determination method that multimedia resource estimates clicking rate, the side
Method includes:
Obtain the user behavior information of user；
The multimedia attribute information of the first multimedia resource is obtained, first multimedia resource is to be recommended to the user's
Multimedia resource；
Clicking rate prediction model is called, the clicking rate prediction model includes embeding layer and clicking rate estimates network, the insertion
Layer includes the corresponding weight matrix of at least one information type, and the clicking rate estimates what network was used to export the embeding layer
Vector is embedded in as input, export multimedia resource estimates clicking rate；
By clicking rate prediction model described in the user behavior information and the multimedia attribute information input, the user is exported
Clicking rate is estimated to first multimedia resource.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811314076.6A CN109408724B (en) | 2018-11-06 | 2018-11-06 | Method and device for determining pre-estimated click rate of multimedia resource and server |
| PCT/CN2019/105452 WO2020093781A1 (en) | 2018-11-06 | 2019-09-11 | Multimedia resource estimated click through rate determination method and apparatus, and server |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811314076.6A CN109408724B (en) | 2018-11-06 | 2018-11-06 | Method and device for determining pre-estimated click rate of multimedia resource and server |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109408724A true CN109408724A (en) | 2019-03-01 |
| CN109408724B CN109408724B (en) | 2020-10-02 |
Family
ID=65471674
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811314076.6A Active CN109408724B (en) | 2018-11-06 | 2018-11-06 | Method and device for determining pre-estimated click rate of multimedia resource and server |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN109408724B (en) |
| WO (1) | WO2020093781A1 (en) |
Cited By (25)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110297970A (en) * | 2019-05-28 | 2019-10-01 | 北京达佳互联信息技术有限公司 | Information recommendation model training method and device |
| CN110472071A (en) * | 2019-07-03 | 2019-11-19 | 中移（杭州）信息技术有限公司 | Multimedia file recommendation method, device, the network equipment and storage medium |
| CN110502715A (en) * | 2019-08-28 | 2019-11-26 | 腾讯科技（深圳）有限公司 | Click the prediction technique and device of probability |
| CN110825956A (en) * | 2019-09-17 | 2020-02-21 | 中国平安人寿保险股份有限公司 | Information flow recommendation method and device, computer equipment and storage medium |
| CN110909182A (en) * | 2019-11-29 | 2020-03-24 | 北京达佳互联信息技术有限公司 | Multimedia resource searching method and device, computer equipment and storage medium |
| WO2020093781A1 (en) * | 2018-11-06 | 2020-05-14 | 北京达佳互联信息技术有限公司 | Multimedia resource estimated click through rate determination method and apparatus, and server |
| CN111160959A (en) * | 2019-12-20 | 2020-05-15 | 深圳前海微众银行股份有限公司 | User click conversion estimation method and device |
| CN111191165A (en) * | 2019-12-05 | 2020-05-22 | 腾讯科技（深圳）有限公司 | Data processing method, device, equipment and computer readable storage medium |
| CN111340522A (en) * | 2019-12-30 | 2020-06-26 | 支付宝实验室(新加坡)有限公司 | Resource recommendation method, device, server and storage medium |
| CN112150182A (en) * | 2019-06-28 | 2020-12-29 | 腾讯科技（深圳）有限公司 | Multimedia file pushing method and device, storage medium and electronic device |
| CN112256918A (en) * | 2020-11-17 | 2021-01-22 | 中国计量大学 | Short video click rate prediction method based on multi-mode dynamic routing |
| CN112307313A (en) * | 2019-07-31 | 2021-02-02 | 北京达佳互联信息技术有限公司 | Information recommendation method and device and storage medium |
| CN112395504A (en) * | 2020-12-01 | 2021-02-23 | 中国计量大学 | Short video click rate prediction method based on sequence capsule network |
| CN112925924A (en) * | 2019-12-05 | 2021-06-08 | 北京达佳互联信息技术有限公司 | Multimedia file recommendation method and device, electronic equipment and storage medium |
| CN112950266A (en) * | 2021-02-26 | 2021-06-11 | 北京达佳互联信息技术有限公司 | Media resource display method and device |
| CN113254679A (en) * | 2021-06-28 | 2021-08-13 | 北京达佳互联信息技术有限公司 | Multimedia resource recommendation method and device, electronic equipment and storage medium |
| CN113495966A (en) * | 2020-03-18 | 2021-10-12 | 北京达佳互联信息技术有限公司 | Determination method and device of interactive operation information and recommendation system of video |
| CN113516522A (en) * | 2021-09-14 | 2021-10-19 | 腾讯科技（深圳）有限公司 | Media resource recommendation method, and training method and device of multi-target fusion model |
| CN113722588A (en) * | 2021-08-12 | 2021-11-30 | 北京达佳互联信息技术有限公司 | Resource recommendation method and device and electronic equipment |
| CN113722578A (en) * | 2020-05-25 | 2021-11-30 | 北京沃东天骏信息技术有限公司 | Data processing method and device |
| CN113742567A (en) * | 2020-05-29 | 2021-12-03 | 北京达佳互联信息技术有限公司 | Multimedia resource recommendation method and device, electronic equipment and storage medium |
| CN114077701A (en) * | 2020-08-13 | 2022-02-22 | 北京达佳互联信息技术有限公司 | Method and device for determining resource information, computer equipment and storage medium |
| CN114547417A (en) * | 2022-02-25 | 2022-05-27 | 北京百度网讯科技有限公司 | Media resource ordering method and electronic equipment |
| CN114677168A (en) * | 2022-03-11 | 2022-06-28 | 百果园技术(新加坡)有限公司 | Resource recommendation method, device, equipment and medium |
| CN114780843A (en) * | 2022-04-20 | 2022-07-22 | 北京字节跳动网络技术有限公司 | Model training method, medium resource recommendation method, device, medium and equipment |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111708944A (en) * | 2020-06-17 | 2020-09-25 | 北京达佳互联信息技术有限公司 | Multimedia resource identification method, device, equipment and storage medium |
| CN111737573A (en) * | 2020-06-17 | 2020-10-02 | 北京三快在线科技有限公司 | Resource recommendation method, device, equipment and storage medium |
| CN113836391B (en) * | 2020-06-24 | 2023-07-11 | 北京达佳互联信息技术有限公司 | Resource recommendation method, device, computer equipment and storage medium |
| CN112364184B (en) * | 2020-11-12 | 2024-04-30 | 北京达佳互联信息技术有限公司 | Method, device, server and storage medium for ordering multimedia data |
| CN113538031B (en) * | 2020-11-18 | 2024-09-10 | 腾讯科技（北京）有限公司 | Training of multimedia resource allocation model and multimedia resource allocation method and device |
| CN112732953B (en) * | 2020-12-30 | 2024-04-12 | 上海众源网络有限公司 | Recommendation method, sample analysis device, electronic equipment and storage medium |
| CN112839247B (en) * | 2021-01-04 | 2023-01-20 | 聚好看科技股份有限公司 | Content display method, display device and cloud server |
| CN114548076A (en) * | 2022-02-28 | 2022-05-27 | 深圳市和讯华谷信息技术有限公司 | Intelligent scoring method for content file and related device |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102663617A (en) * | 2012-03-20 | 2012-09-12 | 亿赞普(北京)科技有限公司 | Method and system for prediction of advertisement clicking rate |
| CN104462594A (en) * | 2014-12-29 | 2015-03-25 | 北京奇虎科技有限公司 | Method and device for providing user personalized resource message pushing |
| CN104951965A (en) * | 2015-06-26 | 2015-09-30 | 深圳市腾讯计算机系统有限公司 | Advertisement delivery method and device |
| CN105095964A (en) * | 2015-08-17 | 2015-11-25 | 杭州朗和科技有限公司 | Data processing method and device |
| CN106604068A (en) * | 2016-12-30 | 2017-04-26 | 中广热点云科技有限公司 | Method and system for updating media program |
| CN108304441A (en) * | 2017-11-14 | 2018-07-20 | 腾讯科技（深圳）有限公司 | Network resource recommended method, device, electronic equipment, server and storage medium |
| CN108307207A (en) * | 2017-12-28 | 2018-07-20 | 北京达佳互联信息技术有限公司 | A kind of video pushing method and device |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105183772A (en) * | 2015-08-07 | 2015-12-23 | 百度在线网络技术（北京）有限公司 | Release information click rate estimation method and apparatus |
| CN105678335B (en) * | 2016-01-08 | 2019-07-02 | 车智互联（北京）科技有限公司 | It estimates the method, apparatus of clicking rate and calculates equipment |
| CN107463580B (en) * | 2016-06-06 | 2020-07-24 | 腾讯科技（深圳）有限公司 | Click rate estimation model training method and device and click rate estimation method and device |
| CN105975641A (en) * | 2016-07-15 | 2016-09-28 | 合网络技术（北京）有限公司 | Video recommendation method ad device |
| CN109408724B (en) * | 2018-11-06 | 2020-10-02 | 北京达佳互联信息技术有限公司 | Method and device for determining pre-estimated click rate of multimedia resource and server |
-
2018
- 2018-11-06 CN CN201811314076.6A patent/CN109408724B/en active Active
-
2019
- 2019-09-11 WO PCT/CN2019/105452 patent/WO2020093781A1/en active Application Filing
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102663617A (en) * | 2012-03-20 | 2012-09-12 | 亿赞普(北京)科技有限公司 | Method and system for prediction of advertisement clicking rate |
| CN104462594A (en) * | 2014-12-29 | 2015-03-25 | 北京奇虎科技有限公司 | Method and device for providing user personalized resource message pushing |
| CN104951965A (en) * | 2015-06-26 | 2015-09-30 | 深圳市腾讯计算机系统有限公司 | Advertisement delivery method and device |
| CN105095964A (en) * | 2015-08-17 | 2015-11-25 | 杭州朗和科技有限公司 | Data processing method and device |
| CN106604068A (en) * | 2016-12-30 | 2017-04-26 | 中广热点云科技有限公司 | Method and system for updating media program |
| CN108304441A (en) * | 2017-11-14 | 2018-07-20 | 腾讯科技（深圳）有限公司 | Network resource recommended method, device, electronic equipment, server and storage medium |
| CN108307207A (en) * | 2017-12-28 | 2018-07-20 | 北京达佳互联信息技术有限公司 | A kind of video pushing method and device |
Non-Patent Citations (1)
| Title |
|---|
| 龙心尘、寒小阳: "从FM推演各深度CTR预估模型(附代码)", 《HTTPS://WWW.JIQIZHIXIN.COM/ARTICLES/2018-07-16-17》 * |
Cited By (38)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020093781A1 (en) * | 2018-11-06 | 2020-05-14 | 北京达佳互联信息技术有限公司 | Multimedia resource estimated click through rate determination method and apparatus, and server |
| CN110297970B (en) * | 2019-05-28 | 2021-04-23 | 北京达佳互联信息技术有限公司 | Information recommendation model training method and device |
| CN110297970A (en) * | 2019-05-28 | 2019-10-01 | 北京达佳互联信息技术有限公司 | Information recommendation model training method and device |
| CN112150182B (en) * | 2019-06-28 | 2023-08-29 | 腾讯科技（深圳）有限公司 | Multimedia file pushing method and device, storage medium and electronic device |
| CN112150182A (en) * | 2019-06-28 | 2020-12-29 | 腾讯科技（深圳）有限公司 | Multimedia file pushing method and device, storage medium and electronic device |
| CN110472071A (en) * | 2019-07-03 | 2019-11-19 | 中移（杭州）信息技术有限公司 | Multimedia file recommendation method, device, the network equipment and storage medium |
| CN112307313B (en) * | 2019-07-31 | 2024-04-16 | 北京达佳互联信息技术有限公司 | Information recommendation method, device and storage medium |
| CN112307313A (en) * | 2019-07-31 | 2021-02-02 | 北京达佳互联信息技术有限公司 | Information recommendation method and device and storage medium |
| CN110502715A (en) * | 2019-08-28 | 2019-11-26 | 腾讯科技（深圳）有限公司 | Click the prediction technique and device of probability |
| CN110502715B (en) * | 2019-08-28 | 2023-07-14 | 腾讯科技（深圳）有限公司 | Click probability prediction method and device |
| CN110825956A (en) * | 2019-09-17 | 2020-02-21 | 中国平安人寿保险股份有限公司 | Information flow recommendation method and device, computer equipment and storage medium |
| CN110909182A (en) * | 2019-11-29 | 2020-03-24 | 北京达佳互联信息技术有限公司 | Multimedia resource searching method and device, computer equipment and storage medium |
| CN112925924A (en) * | 2019-12-05 | 2021-06-08 | 北京达佳互联信息技术有限公司 | Multimedia file recommendation method and device, electronic equipment and storage medium |
| CN111191165A (en) * | 2019-12-05 | 2020-05-22 | 腾讯科技（深圳）有限公司 | Data processing method, device, equipment and computer readable storage medium |
| CN111160959B (en) * | 2019-12-20 | 2023-06-27 | 深圳前海微众银行股份有限公司 | User click conversion prediction method and device |
| CN111160959A (en) * | 2019-12-20 | 2020-05-15 | 深圳前海微众银行股份有限公司 | User click conversion estimation method and device |
| CN111340522B (en) * | 2019-12-30 | 2024-03-08 | 支付宝实验室(新加坡)有限公司 | Resource recommendation method, device, server and storage medium |
| CN111340522A (en) * | 2019-12-30 | 2020-06-26 | 支付宝实验室(新加坡)有限公司 | Resource recommendation method, device, server and storage medium |
| CN113495966A (en) * | 2020-03-18 | 2021-10-12 | 北京达佳互联信息技术有限公司 | Determination method and device of interactive operation information and recommendation system of video |
| CN113495966B (en) * | 2020-03-18 | 2023-06-23 | 北京达佳互联信息技术有限公司 | Interactive operation information determining method and device and video recommendation system |
| CN113722578A (en) * | 2020-05-25 | 2021-11-30 | 北京沃东天骏信息技术有限公司 | Data processing method and device |
| CN113742567B (en) * | 2020-05-29 | 2023-08-22 | 北京达佳互联信息技术有限公司 | Recommendation method and device for multimedia resources, electronic equipment and storage medium |
| CN113742567A (en) * | 2020-05-29 | 2021-12-03 | 北京达佳互联信息技术有限公司 | Multimedia resource recommendation method and device, electronic equipment and storage medium |
| CN114077701A (en) * | 2020-08-13 | 2022-02-22 | 北京达佳互联信息技术有限公司 | Method and device for determining resource information, computer equipment and storage medium |
| CN112256918A (en) * | 2020-11-17 | 2021-01-22 | 中国计量大学 | Short video click rate prediction method based on multi-mode dynamic routing |
| CN112256918B (en) * | 2020-11-17 | 2021-07-06 | 中国计量大学 | Short video click rate prediction method based on multi-mode dynamic routing |
| CN112395504B (en) * | 2020-12-01 | 2021-11-23 | 中国计量大学 | Short video click rate prediction method based on sequence capsule network |
| CN112395504A (en) * | 2020-12-01 | 2021-02-23 | 中国计量大学 | Short video click rate prediction method based on sequence capsule network |
| CN112950266B (en) * | 2021-02-26 | 2024-05-03 | 北京达佳互联信息技术有限公司 | Media resource display method and device |
| CN112950266A (en) * | 2021-02-26 | 2021-06-11 | 北京达佳互联信息技术有限公司 | Media resource display method and device |
| CN113254679B (en) * | 2021-06-28 | 2021-09-28 | 北京达佳互联信息技术有限公司 | Multimedia resource recommendation method and device, electronic equipment and storage medium |
| CN113254679A (en) * | 2021-06-28 | 2021-08-13 | 北京达佳互联信息技术有限公司 | Multimedia resource recommendation method and device, electronic equipment and storage medium |
| CN113722588B (en) * | 2021-08-12 | 2023-09-05 | 北京达佳互联信息技术有限公司 | Resource recommendation method and device and electronic equipment |
| CN113722588A (en) * | 2021-08-12 | 2021-11-30 | 北京达佳互联信息技术有限公司 | Resource recommendation method and device and electronic equipment |
| CN113516522A (en) * | 2021-09-14 | 2021-10-19 | 腾讯科技（深圳）有限公司 | Media resource recommendation method, and training method and device of multi-target fusion model |
| CN114547417A (en) * | 2022-02-25 | 2022-05-27 | 北京百度网讯科技有限公司 | Media resource ordering method and electronic equipment |
| CN114677168A (en) * | 2022-03-11 | 2022-06-28 | 百果园技术(新加坡)有限公司 | Resource recommendation method, device, equipment and medium |
| CN114780843A (en) * | 2022-04-20 | 2022-07-22 | 北京字节跳动网络技术有限公司 | Model training method, medium resource recommendation method, device, medium and equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2020093781A1 (en) | 2020-05-14 |
| CN109408724B (en) | 2020-10-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109408724A (en) | Multimedia resource estimates the determination method, apparatus and server of clicking rate | |
| WO2020011068A1 (en) | Method and system for executing machine learning process | |
| US7774349B2 (en) | Statistical models and methods to support the personalization of applications and services via consideration of preference encodings of a community of users | |
| CN109062919B (en) | Content recommendation method and device based on deep reinforcement learning | |
| CN109992710A (en) | Clicking rate predictor method, system, medium and calculating equipment | |
| US10949000B2 (en) | Sticker recommendation method and apparatus | |
| JP7564332B2 (en) | Method and device for identifying fusion parameters, method and device for recommending information, method and device for training a parameter measurement model, electronic device, storage medium, and computer program | |
| US10019491B1 (en) | Machine learning of response selection to structured data input | |
| CN110413894B (en) | Training method of content recommendation model, content recommendation method and related device | |
| US20170017649A1 (en) | Techniques for providing a user with content recommendations | |
| US10474926B1 (en) | Generating artificial intelligence image processing services | |
| CN109558514A (en) | Video recommendation method, its device, information processing equipment and storage medium | |
| US11126785B1 (en) | Artificial intelligence system for optimizing network-accessible content | |
| US20220414470A1 (en) | Multi-Task Attention Based Recurrent Neural Networks for Efficient Representation Learning | |
| CN113222073A (en) | Method and device for training transfer learning model and recommendation model | |
| CN113420165B (en) | Training of classification model and classification method and device of multimedia data | |
| CN110531951A (en) | Picture output mode method of adjustment, device, equipment, system and storage medium | |
| CN109978575A (en) | A kind of method and device excavated customer flow and manage scene | |
| CN110377821A (en) | Generate method, apparatus, computer equipment and the storage medium of interest tags | |
| CN112115354B (en) | Information processing method, device, server and storage medium | |
| CN112269943B (en) | Information recommendation system and method | |
| CN109446379A (en) | Method and apparatus for handling information | |
| CN113836388A (en) | Information recommendation method and device, server and storage medium | |
| JP2014130520A (en) | Method, computer system, and computer program for optimizing scheme for selecting action maximizing expectation return while suppressing risk | |
| CN113935251A (en) | User behavior prediction model generation method and device and user behavior prediction method and device |
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 |