CN107222787A - Video resource popularity prediction method - Google Patents
Video resource popularity prediction method Download PDFInfo
- Publication number
- CN107222787A CN107222787A CN201710409198.2A CN201710409198A CN107222787A CN 107222787 A CN107222787 A CN 107222787A CN 201710409198 A CN201710409198 A CN 201710409198A CN 107222787 A CN107222787 A CN 107222787A
- Authority
- CN
- China
- Prior art keywords
- mrow
- msub
- data
- rating
- mover
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
- H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
- H04N21/45—Management operations performed by the client for facilitating the reception of or the interaction with the content or administrating data related to the end-user or to the client device itself, e.g. learning user preferences for recommending movies, resolving scheduling conflicts
- H04N21/466—Learning process for intelligent management, e.g. learning user preferences for recommending movies
- H04N21/4667—Processing of monitored end-user data, e.g. trend analysis based on the log file of viewer selections
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/70—Information retrieval; Database structures therefor; File system structures therefor of video data
- G06F16/73—Querying
- G06F16/738—Presentation of query results
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
- H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
- H04N21/45—Management operations performed by the client for facilitating the reception of or the interaction with the content or administrating data related to the end-user or to the client device itself, e.g. learning user preferences for recommending movies, resolving scheduling conflicts
- H04N21/4508—Management of client data or end-user data
- H04N21/4532—Management of client data or end-user data involving end-user characteristics, e.g. viewer profile, preferences
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Signal Processing (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of video resource Popularity prediction method, including：The viewing-data of certain area in-group user is counted, the rating categorical data and interbehavior data of group of subscribers is obtained, and utilizes the resource popularity of rating categorical data statistics to calculate；Utilize the LDA models for coupling user behavior, traversal rating categorical data and interbehavior data generate correspondence Di Li Cray distributions respectively, the full probability of each behavior pattern is derived by chain rule and the expectation of its Di Li Cray distribution is asked for, behavior pattern matrix is obtained；With reference to neural network model, the resource popularity of institute's statistics and behavior pattern matrix are inputted as neutral net, by training generation forecast model, and then to predict the video resource popularity in future.This method considers the influence that user watched interactive contents data and interbehavior data are predicted resource popularity, and the relation between two class data of research and popularity improves the accuracy predicted resource popularity.
Description
Technical field
The present invention relates to data mining and machine learning techniques field, a kind of video resource Popularity prediction side is especially designed
Method.
Background technology
Video request program is a kind of based on user watched dynamic need, and the service content for transmitting and playing corresponding demand is regarded
Frequency play-back technology.User has the initiative to data flow in the selection of rating content and viewing mode, if can ring in time
Efficiency using family viewing behavior demand and scheduling of resource is closely related.During scheduling of resource, the popularity of resource information
It is the important references standard of dispatching algorithm.
Separately below related research work is introduced from user behavior analysis and resource popularity forecasting research respectively.
In the existing research on user behavior analysis, the definition to user behavior is broadly divided into two classes：What one class referred to
It is interactive user operations of the user during program request rating, described interactive operation and the particular problem of research has closely
The system of " user behavior analysis device and realization in IPTV system " analysis application of this network technology company of relation, such as Saite,
How main research is safeguarded and be run to equipment control model with reference to user behavior, and different user data informations is counted
Service provider is fed back to after arrangement, user behavior therein generally refers to user to live, program request, the use feelings reviewed
Condition, including viewing duration and viewing number of times etc., user behavior therein are fixed under its company oneself equipment management system framework
Justice is simultaneously counted, and the difference between viewing mode (live, program request) is confined to the definition of viewing behavior, without it is real deeply
To in demand mode in different user mutual behaviors；Another kind of user behavior refers to the content of user's request, to ask
Content represent user behavior and user's request content handled and excavated, such as Wang Pan patent " is regarded based on full-service
The IPTV user behavior analysis method of map analysis " matches the customer group square for being adapted to preference according to service attribute centered on business
Battle array, user behavior therein generally refers to demand of the user to IPTV value-added services, including information browse, game, visual electricity
The value-added services such as words, image space, think to its study limitation to user behavior that use of the user to different business is represented not
Same user behavior, is not still deep into the essence of user mutual behavior, it is impossible to react user exactly to rating content
Demand.
For resource popularity prediction, " a kind of Resource service system and its resource allocation methods " of CHINAUNICOM mainly will
Queued tasks in operation system carry out Clustering according to weight properties, and resource point is carried out to task according to priority marking
Match somebody with somebody；Four reach " Proxy Caching for Streaming Media replacement method and the device " of epoch communication network technique Co., Ltd mainly according to principal component
The method of analysis and multiple linear regression determines the Popularity prediction value of Streaming Media Object, is determined with reference to the signal to noise ratio of video peak value
The comprehensive value of Streaming Media Object is to carry out caching replacement；Zhang Tiankui etc. is " in a kind of information predicted based on content popularit
Proposed in heart network-caching method " and clustered each nodes records by similarity analysis, the content in calculate node
Popularity simultaneously does cycle statistics, is predicted further according to prediction algorithm pop degree, and carrying out contrast decision with local packets is
No renewal nodal cache；Often mark etc. in " a kind of online serialized content Popularity prediction method based on autoregression model "
The prediction of online serialized content popularity is solved again, and by crawling the overall playback volume trend of online serialized content, parsing is overall
The html source code of the playback volume trend page, utilizes the popularity of the new serialized content of autoregressive model prediction.Come from such scheme
See, the prediction of popularity and user mutual behavior are not carried out Conjoint Analysis by existing research so that the forecasting research of popularity
The data supporting of user behavior angle is lacked, Popularity prediction precision can not obtain Reliable guarantee.
The content of the invention
It is an object of the invention to provide a kind of video resource Popularity prediction method, user watched interaction content is considered
The influence that data and interbehavior data are predicted resource popularity, the relation between two class data of research and popularity is improved
The accuracy predicted resource popularity.
The purpose of the present invention is achieved through the following technical solutions：
A kind of video resource Popularity prediction method, including：
The viewing-data of certain area in-group user is counted, the rating categorical data and interbehavior of group of subscribers is obtained
Data, and utilize the resource popularity of rating categorical data statistics to calculate；
Using the document subject matter generation model LDA for coupling user behavior, traversal rating categorical data and interbehavior data
Correspondence Di Li Cray distributions are generated respectively, and the full probability of each behavior pattern is derived by chain rule and its Di Li Cray is asked for
The expectation of distribution, obtains behavior pattern matrix；
It is with reference to neural network model, the resource popularity of institute's statistics and behavior pattern matrix is defeated as neutral net
Enter, by training generation forecast model, the video resource popularity in future is predicted according to the forecast model of training.
The viewing-data of the statistics certain area in-group user, obtains rating categorical data and the interaction of group of subscribers
Behavioral data includes：
The viewing daily record of all group of subscribers in area is counted, by dividing period and program category, colony is obtained
The rating categorical data and the data of interbehavior of different time sections in the certain number of days of user；
Wherein, the division period refers to divide the time of one day, if being divided into 24 periods by one day, often
One period was a hour；
Interbehavior refers to broadcast state of the user in demand (telecommunication) service, has 10 kinds：Program collection, first startup are broadcast
Put, be again started up broadcasting, F.F. state, fast-rewinding state, halted state, positioning playing, play failure, exit to play and broadcast with memory
Put.
The resource popularity using rating categorical data statistics to calculate includes：
By rating categorical data set C={ ci| i=1,2, K } represent, the element c in set CiBy rating
Type number, ciI class rating types are represented, accounting of its time being accessed by the user in total rating duration is pi, then piIt is ci
Popularity in set C, and P={ pi| i=1,2, K be set C resource popularity set；Wherein, K is receipts
Depending on the total quantity of type.
It is described to utilize the document subject matter generation model LDA for coupling user behavior, traversal rating categorical data and interbehavior
Data generate correspondence Di Li Cray distributions respectively, derive the full probability of each behavior pattern by chain rule and ask for its Di Li
The expectation of Cray distribution, obtaining behavior pattern matrix includes：
Assuming that there is K kind rating types V in rating categorical datawThere is L kind behavior types V in portion's program, interbehavior datal
Individual interbehavior；
To there is the interbehavior of a certain type in the program of a certain rating type, then referred to as behavior pattern；M-th of document
The behavior pattern multinomial distribution of middle userIt is K × L dimension matrix, behavior pattern distributionIn z-th positional representation
Behavior pattern is user to rating type zmn1Program take zmn2Plant interbehavior；Wherein,
zmn2=(zmnmod K)
Using the LDA models for coupling user behavior, traversal rating categorical data and interbehavior data generate correspondence respectively
Di Li Crays are distributed, and its process is as follows：
Order：
Then for each viewing-data n, n ∈ { 1 ..., Nm}；Wherein, NmFor viewing-data quantity in m-th of document,For kth kind
The program multinomial distribution of rating type,For the interbehavior multinomial distribution of l kind behavior types, α is that behavior pattern is more
Item formula distributionDirichlet Study firsts, β be program multinomial distributionDirichlet Study firsts, γ for interaction
Behavior multinomial distributionDirichlet Study firsts；
Calculate the distribution of Di Li Crays Θ, Φ, Ψ：
Wherein,Behavior pattern, program category, the quantity of behavior type for participating in calculating are represented respectively；Respectively program, interbehavior and behavior pattern set；Represent the set of behavior pattern in m-th of document, Φ
For program profile multinomialIn all program probability of happening Di Li Crays distribution, Ψ be interbehavior multinomial distributionIn
The Di Li Crays distribution of all interbehavior probability of happening, Θ is behavior pattern multinomial distributionIn all behavior patterns occur
The Di Li Crays distribution of probability；
Θ, Φ, Ψ are distributed based on Di Li Crays, and derive by chain rule the full probability of each behavior pattern：
Wherein,Respectively program, interbehavior and behavior pattern set, these three set inner elements wmn、
tmn、zmnRepresent respectively the program of the n-th viewing-data in m-th of document, the interbehavior of the n-th viewing-data in m-th of document,
The behavior pattern of n-th viewing-data in m-th of document；Represent behavior pattern setMiddle removal zmnSet afterwards；
The full probability for being based ultimately upon Di Li Crays distribution Θ, Φ, Ψ and each behavior pattern solves the distribution of Di Li Crays
Expectation, be derived by equation below：
Wherein,User is represented to V in kth kind rating typewThe viewing number of times of portion's program,Distribution is represented in l
Plant the V of behavior typelThe number of times of interbehavior,Represent the frequency of i-th of behavior pattern in m-th of document；θmi、Respectively single behavior pattern, program, the probability of happening of interbehavior.
The combination neural network model, regard the resource popularity of institute's statistics and behavior pattern matrix as nerve net
Network is inputted, and by training generation forecast model, the video resource popularity in future is predicted according to the forecast model of training to be included：
The Nonlinear Mapping to behavior pattern matrix and the resource popularity of statistics is realized using BP neural network, it is defeated
Enter layer and output layer neuron number is relevant with input/output argument, the rating type of the behavior pattern matrix program of input has K
Kind, the video resource popularity in future is output as, then input layer and output layer node number are respectively K and 1, node in hidden layer
It is set to S；
By the behavior vector of the program of different rating types in the resource popularity of statistics, and behavior pattern matrix
Read in, then training data and test data are divided into by data are read in；
Start to initialize BP neural network, training method uses steepest descent method, and batch mode trains behavioral pattern data,
Then sample is inputted into BP neural network by the way of batch is trained, calculates the error of each sample；Finally judge whether to receive
Hold back, if not restraining, weights are adjusted according to steepest descent method, until convergence, so as to obtain forecast model；
The test data of delimitation is inputted into BP neural network, predicts that the video resource in future is popular using forecast model
Degree.
As seen from the above technical solution provided by the invention, colony is carried out by program category to user data in area
Analysis, effectively weakens the influence that single film breaks out pop degree precision of prediction in short term；By the LDA moulds for coupling user behavior
Type carries out Conjoint Analysis to rating content and interbehavior, it is to avoid loss of learning caused by the single data of research, more accurately
Ground finds group of subscribers behavior pattern；Using neural network model, according to the behavior pattern Matrix prediction for having merged behavioral data
Video resource popularity, improves the Nonlinear Processing ability of prediction, so as to reduce predicated error；In addition, the present invention is provided
Method be applied in the scheduling of resource of Cloud Server, can effectively improve request receptance, lift Consumer's Experience.
Brief description of the drawings
In order to illustrate the technical solution of the embodiments of the present invention more clearly, being used required in being described below to embodiment
Accompanying drawing be briefly described, it should be apparent that, drawings in the following description are only some embodiments of the present invention, for this
For the those of ordinary skill in field, on the premise of not paying creative work, other can also be obtained according to these accompanying drawings
Accompanying drawing.
Fig. 1 is a kind of flow chart of video resource Popularity prediction method provided in an embodiment of the present invention；
Fig. 2 is the flow chart that behavior pattern provided in an embodiment of the present invention is found；
Fig. 3 is the schematic diagram of behavior pattern matrix provided in an embodiment of the present invention；
Fig. 4 is the flow chart of the video resource Popularity prediction of fusion user behavior provided in an embodiment of the present invention.
Embodiment
With reference to the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete
Ground is described, it is clear that described embodiment is only a part of embodiment of the invention, rather than whole embodiments.Based on this
The embodiment of invention, the every other implementation that those of ordinary skill in the art are obtained under the premise of creative work is not made
Example, belongs to protection scope of the present invention.
The embodiment of the present invention provides a kind of video resource Popularity prediction method, and this method is to user group's data from resource
Type dimension is handled, and generates user group's viewing-data；It is not independent handle in rating perhaps for viewing-data analysis
Person's interbehavior, but two class data of fusion derive its joint probability distribution, so as to accurately describe the internal relation of two class data；
In Forecasting Methodology, joint probability Input matrix to neural network model is predicted to video resource popularity, set up between the two
Accurate mapping relations.
As described in Figure 1, it is a kind of flow chart of video resource Popularity prediction method.First, group in statistics certain area
The viewing-data of body user, obtains the rating categorical data and interbehavior data of group of subscribers, and utilizes rating categorical data
The resource popularity of statistics to calculate；Then, using the document subject matter generation model LDA for coupling user behavior, traversal
Rating categorical data and interbehavior data generate correspondence Di Li Cray distributions respectively, and each behavior mould is derived by chain rule
The full probability of formula and the expectation for asking for the distribution of its Di Li Cray, obtain behavior pattern matrix；Finally, with reference to neural network model,
The resource popularity of institute's statistics and behavior pattern matrix are inputted as neutral net, by training generation forecast model,
The video resource popularity in future is predicted according to the forecast model of training.This method considers user watched interaction content number
According to the influence predicted with interbehavior data resource popularity, the relation between two class data of research and popularity, raising pair
The accuracy of resource popularity prediction, and then it is effectively improved the resource deployment efficiency of Streaming Media Cloud Server, lifting request access
With the service quality of response.
It is described in detail below for each step.
First, group of subscribers data processing.
In the embodiment of the present invention, the purpose of group of subscribers data processing is, by counting all group of subscribers in area
Viewing daily record, then by dividing period and program category, different time sections in the certain number of days of group of subscribers can be obtained
The data of rating categorical data and interbehavior.Meanwhile, the behavior mould of individual user can effectively be ignored to population data processing
Formula makes a variation, and behavior pattern variation refers to certain user because with special behavior pattern, such as a F.F. or only sees certain class
Program.Can be by the influence reduction of these uncertain variation pop degree predictions to minimum from colony's angle processing.
In the embodiment of the present invention, the division period refers to divide the time of one day, if being divided into 24 by one day
Period, then each period is a hour, then the data set sample of 25 days can just be divided into 600 period collection
Close.
Rating type can be divided into 15 types, respectively news, finance and economics, variety, physical culture, film, animation, military affairs,
TV play, science and education, life, fashion tourism, child-parent education, music, old man's program and juvenile's program.
Interbehavior refers to broadcast state of the user in demand (telecommunication) service, has 10 kinds：Program collection, first startup are broadcast
Put, be again started up broadcasting, F.F. state, fast-rewinding state, halted state, positioning playing, play failure, exit to play and broadcast with memory
Put.
In addition, the embodiment of the present invention is also using the resource popularity of rating categorical data statistics to calculate, as
One input of neutral net.Specifically：By rating categorical data set C={ ci| i=1,2, K } represent,
Element c in set CiBy rating type number, ciI class rating types are represented, its time being accessed by the user is in total rating duration
In accounting be pi, then piIt is ciPopularity in set C, and P={ pi| i=1,2, K be set C resource
Popularity set；Wherein, K is the total quantity of rating type.
2nd, behavior pattern is found.
LDA models can obtain a data and concentrate complicated theme as classical body of text generation model
Set, under broadcasting and TV data research background, the present invention is ground using the LDA models of coupling user behavior to group of subscribers data
Study carefully.As shown in Fig. 2 choosing the viewing-data in user's one period of past, type demarcation is carried out to each program, program
Type is as the theme of document generation, and LDA models can accurately generate the set of different program categories；Different program classes
The viewing behavior pattern of user is different in type, and therefore, behavior pattern is distributedIntroduce LDA models to be coupled, by row
User watched program category is finely divided for pattern, can be so in the rating to group of subscribers in finer granularity
Appearance is analysed in depth, and obtains the Joint Distribution of behavior-program, behavior pattern matrix can be obtained in conjunction with subsequent treatment.
The LDA models of user behavior are coupled as the extension of LDA models, Conjoint Analysis can be carried out to behavioral agent.Knot
Close Fig. 3, it is assumed that have K kind rating types V in rating categorical datawPortion's program, then the Joint Distribution Φ ' of type-program is K × Vw
Tie up matrix,Represent that K kind ratings type includes VwThe probability of portion's program；Assuming that there is L kind behavior classes in interbehavior data
Type VlIndividual interbehavior, then the Joint Distribution Ψ ' of type-behavior is L × VlTie up matrix,Represent in l kind behavior types
VlThe probability of individual interbehavior；
In the embodiment of the present invention, there is the interbehavior of a certain type for the program of a certain rating type, referred to as go
For pattern；The behavior pattern multinomial distribution of user in m-th of documentIt is K × L dimension matrix, behavior pattern distributionIn
The behavior pattern of z-th of positional representation is user to rating type zmn1Program take zmn2Plant interbehavior；Wherein,
zmn2=(zmnmod K)
Using the LDA models for coupling user behavior, traversal rating categorical data and interbehavior data generate correspondence respectively
Di Li Crays are distributed, and its process is as follows：
Order：
Then for each viewing-data n (including timestamp, programm name, behavior type in rating record), n ∈ { 1 ..., Nm}；
Wherein, NmFor viewing-data quantity in m-th of document,For the program multinomial distribution of kth kind rating type,For l kinds
The interbehavior multinomial distribution of behavior type, α is behavior pattern multinomial distributionDirichlet Study firsts, β for section
Mesh multinomial distributionDirichlet Study firsts, γ be interbehavior multinomial distributionDirichlet Study firsts；
These multinomial distributions are made up of multiple elements, for example, program multinomial distributionK-th of element beTwo other many item number
The implication of distributed constant is similar.
Calculate Di Li Crays distribution Θ：
In above formula,Represent to consider Study first and behavior time in Di Li Crays distribution pilot process is calculated
Several probability.
The distribution of Di Li Crays Φ, Ψ can similarly be obtained：
Wherein,The behavior pattern, rating type, the quantity of behavior type for participating in calculating are represented respectively；Respectively program, interbehavior and behavior pattern set；Represent the set of behavior pattern in m-th of document, Φ
For program profile multinomialIn all program probability of happening Di Li Crays distribution, Ψ be interbehavior multinomial distributionIn
The Di Li Crays distribution of all interbehavior probability of happening, Θ is behavior pattern multinomial distributionIn all behavior patterns occur
The Di Li Crays distribution of probability.
Θ, Φ, Ψ are distributed based on Di Li Crays, and derive by chain rule the full probability of each behavior pattern：
Wherein,Three set inner element wmn、tmn、zmnNth bar rating number in m-th of document is represented respectively
According to program, in m-th of document in the interbehavior of nth bar viewing-data, m-th of document nth bar viewing-data behavior mould
Formula；Represent behavior pattern setMiddle removal zmnSet afterwards；
The implication of above formula refers to Study first (i.e. α, β, γ) and program, behavior and the behavior being distributed according to existing three class
The set of pattern is (i.e.) ask for the full probability that each behavior pattern occurs.The calculating of molecule and denominator on the right of formula
It is respectively necessary for using Φ, Ψ, Θ in journey, that is,
In addition,Signal table it is as shown in table 1.
Table 1Signal table
The first row first row θ in upper table11To last row of last column θMIThe probability of happening combination of all behavior patterns
ConstituteNamely a behavior pattern matrix.The probability of happening of any one behavior pattern therein is designated as θmi, it is also final herein
Want the result asked for.
The full probability for being based ultimately upon Di Li Crays distribution Θ, Φ, Ψ and each behavior pattern solves the distribution of Di Li Crays
Expectation, be derived by equation below：
Wherein,User is represented to V in kth kind rating typewThe viewing number of times of portion's program,Distribution is represented in l
Plant the V of behavior typelThe number of times of interbehavior,Represent the frequency of i-th of behavior pattern in m-th of document；θmi、Respectively single behavior pattern, program, the probability of happening of interbehavior.In the embodiment of the present invention, by Di Li
The expectation of Cray distribution is solved, and obtains θmi、It is therein, θmiNamely represent different elements in behavior pattern matrix
Probability of happening, be our output.
In the embodiment of the present invention, the implication explanation of parameters is as shown in table 2.
The cLDA Model Parameters of the fusion user behavior of table 2 illustrate table
3rd, the video resource Popularity prediction of user behavior is merged
The video resource Popularity prediction of fusion user behavior refers to, is predicted, will merged using neural network model
The behavior pattern Input matrix models of user watched data and interbehavior, makes each behavior pattern matrix be mapped to one
Popularity set, according to model prediction future video resource popularity.
In the embodiment of the present invention, realized using BP neural network to behavior pattern matrix and the resource popularity of statistics
Nonlinear Mapping, input layer and output layer neuron number be relevant with input/output argument, the behavior pattern matrix section of input
Purpose rating type has K kinds, is output as the video resource popularity in future, then input layer and output layer node number are respectively K
With 1, node in hidden layer is set to S.
As shown in figure 4, by the program of different rating type in the resource popularity of statistics, and behavior pattern matrix
Behavior vector read in, then by read in data be divided into training data and test data；Start to initialize BP neural network, training
Method uses steepest descent method, and then batch mode training behavioral pattern data inputs sample by the way of batch is trained
BP neural network, calculates the error of each sample, exemplary, and the error margin that can define popularity set is 0.1；Finally
Judge whether convergence, if not restraining, weights are adjusted according to steepest descent method, until convergence, so as to obtain forecast model；It will delimit
Test data input BP neural network, predicted using forecast model future video resource popularity.Finally, it will measure in advance
To video resource popularity and the obtained video popularity of statistics apply respectively on Cloud Server, it is recognised that of the invention
There is significant actively impact to performance boost.
Such scheme of the embodiment of the present invention, population analysis, effectively reduction are carried out to user data in area by program category
Single film breaks out the influence of pop degree precision of prediction in short term；By coupling the LDA models of user behavior to rating content
Conjoint Analysis is carried out with interbehavior, it is to avoid loss of learning caused by research single data, more accurately find that colony uses
Family behavior pattern；It is popular according to the behavior pattern Matrix prediction video resource for having merged behavioral data using neural network model
Degree, improves the Nonlinear Processing ability of prediction, so as to reduce predicated error；In addition, the method that the present invention is provided is applied to
In the scheduling of resource of Cloud Server, request receptance can be effectively improved, Consumer's Experience is lifted.
Through the above description of the embodiments, those skilled in the art can be understood that above-described embodiment can
To be realized by software, the mode of necessary general hardware platform can also be added to realize by software.Understood based on such,
The technical scheme of above-described embodiment can be embodied in the form of software product, the software product can be stored in one it is non-easily
The property lost storage medium (can be CD-ROM, USB flash disk, mobile hard disk etc.) in, including some instructions are to cause a computer to set
Standby (can be personal computer, server, or network equipment etc.) performs the method described in each embodiment of the invention.
The foregoing is only a preferred embodiment of the present invention, but protection scope of the present invention be not limited thereto,
Any one skilled in the art is in the technical scope of present disclosure, the change or replacement that can be readily occurred in,
It should all be included within the scope of the present invention.Therefore, protection scope of the present invention should be with the protection model of claims
Enclose and be defined.
Claims (5)
1. a kind of video resource Popularity prediction method, it is characterised in that including：
The viewing-data of certain area in-group user is counted, the rating categorical data and interbehavior number of group of subscribers is obtained
According to, and utilize the resource popularity of rating categorical data statistics to calculate；
Using the document subject matter generation model LDA for coupling user behavior, traversal rating categorical data and interbehavior data difference
Generation correspondence Di Li Cray distributions, derive the full probability of each behavior pattern by chain rule and ask for the distribution of its Di Li Cray
Expectation, obtain behavior pattern matrix；
With reference to neural network model, the resource popularity of institute's statistics and behavior pattern matrix are inputted as neutral net,
By training generation forecast model, the video resource popularity in future is predicted according to the forecast model of training.
2. a kind of video resource Popularity prediction method according to claim 1, it is characterised in that the certain area of statistics
The viewing-data of domain in-group user, obtaining the rating categorical data and interbehavior data of group of subscribers includes：
The viewing daily record of all group of subscribers in area is counted, by dividing period and program category, group of subscribers is obtained
The rating categorical data and the data of interbehavior of different time sections in certain number of days；
Wherein, the division period refers to divide the time of one day, if being divided into 24 periods by one day, per for the moment
Between section be a hour；
Interbehavior refers to broadcast state of the user in demand (telecommunication) service, has 10 kinds：Program collection, first startup are played, again
Secondary startup broadcasting, F.F. state, fast-rewinding state, halted state, positioning playing, broadcasting fail, exit broadcasting and memory broadcasting.
3. a kind of video resource Popularity prediction method according to claim 1, it is characterised in that the utilization rating class
The resource popularity of type data statistics to calculate includes：
By rating categorical data set C={ ci| i=1,2 ..., K } represent, the element c in set CiBy rating type number,
ciI class rating types are represented, accounting of its time being accessed by the user in total rating duration is pi, then piIt is ciIn set C
Popularity, and P={ pi| i=1,2 ..., K be set C resource popularity set；Wherein, K is the sum of rating type
Amount.
4. a kind of video resource Popularity prediction method according to claim 1, it is characterised in that described to be used using coupling
The document subject matter generation model LDA of family behavior, traversal rating categorical data and interbehavior data generate correspondence Di Like respectively
Thunder is distributed, and derives the full probability of each behavior pattern by chain rule and asks for the expectation of its Di Li Cray distribution, OK
Include for mode matrix：
Assuming that there is K kind rating types V in rating categorical datawThere is L kind behavior types V in portion's program, interbehavior datalIt is individual to hand over
Mutual behavior；
To there is the interbehavior of a certain type in the program of a certain rating type, then referred to as behavior pattern；Used in m-th of document
The behavior pattern multinomial distribution at familyIt is K × L dimension matrix, behavior pattern distributionIn z-th of positional representation behavior
Pattern is user to rating type zmn1Program take zmn2Plant interbehavior；Wherein,
<mrow>
<msub>
<mi>z</mi>
<mrow>
<mi>m</mi>
<mi>n</mi>
<mn>1</mn>
</mrow>
</msub>
<mo>=</mo>
<mrow>
<mo>(</mo>
<mfrac>
<mrow>
<msub>
<mi>z</mi>
<mrow>
<mi>m</mi>
<mi>n</mi>
</mrow>
</msub>
<mo>-</mo>
<mn>1</mn>
</mrow>
<mi>L</mi>
</mfrac>
<mo>+</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
<mo>;</mo>
</mrow>
zmn2=(zmnmod K)
Using the LDA models for coupling user behavior, traversal rating categorical data and interbehavior data generate correspondence Di Li respectively
Cray is distributed, and its process is as follows：
Order：Then
For each viewing-data n, n ∈ { 1 ..., Nm}；Wherein, NmFor viewing-data quantity in m-th of document,Received for kth kind
Depending on the program multinomial distribution of type,For the interbehavior multinomial distribution of l kind behavior types, α is that behavior pattern is multinomial
Formula is distributedDirichlet Study firsts, β be program multinomial distributionDirichlet Study firsts, γ be interaction row
For multinomial distributionDirichlet Study firsts；
Calculate the distribution of Di Li Crays Θ, Φ, Ψ：
<mrow>
<mi>P</mi>
<mrow>
<mo>(</mo>
<msub>
<mover>
<mi>&theta;</mi>
<mo>&RightArrow;</mo>
</mover>
<mi>m</mi>
</msub>
<mo>|</mo>
<msub>
<mover>
<mi>z</mi>
<mo>&RightArrow;</mo>
</mover>
<mrow>
<mi>m</mi>
<mo>,</mo>
</mrow>
</msub>
<mi>&alpha;</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mi>D</mi>
<mi>i</mi>
<mi>r</mi>
<mi>i</mi>
<mi>c</mi>
<mi>h</mi>
<mi>l</mi>
<mi>e</mi>
<mi>t</mi>
<mrow>
<mo>(</mo>
<msub>
<mover>
<mi>&theta;</mi>
<mo>&RightArrow;</mo>
</mover>
<mi>m</mi>
</msub>
<mo>|</mo>
<msub>
<mover>
<mi>n</mi>
<mo>&RightArrow;</mo>
</mover>
<mi>m</mi>
</msub>
<mo>+</mo>
<mi>&alpha;</mi>
<mo>)</mo>
</mrow>
<mo>;</mo>
</mrow>
<mrow>
<mi>P</mi>
<mrow>
<mo>(</mo>
<msub>
<mover>
<mi>&phi;</mi>
<mo>&RightArrow;</mo>
</mover>
<mi>k</mi>
</msub>
<mo>|</mo>
<mover>
<mi>z</mi>
<mo>&RightArrow;</mo>
</mover>
<mo>,</mo>
<mover>
<mi>w</mi>
<mo>&RightArrow;</mo>
</mover>
<mo>,</mo>
<mi>&beta;</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mi>D</mi>
<mi>i</mi>
<mi>r</mi>
<mi>i</mi>
<mi>c</mi>
<mi>h</mi>
<mi>l</mi>
<mi>e</mi>
<mi>t</mi>
<mrow>
<mo>(</mo>
<msub>
<mover>
<mi>&phi;</mi>
<mo>&RightArrow;</mo>
</mover>
<mi>k</mi>
</msub>
<mo>|</mo>
<msub>
<mover>
<mi>n</mi>
<mo>&RightArrow;</mo>
</mover>
<mi>k</mi>
</msub>
<mo>+</mo>
<mi>&beta;</mi>
<mo>)</mo>
</mrow>
<mo>;</mo>
</mrow>
<mrow>
<mi>P</mi>
<mrow>
<mo>(</mo>
<msub>
<mover>
<mi>&psi;</mi>
<mo>&RightArrow;</mo>
</mover>
<mi>l</mi>
</msub>
<mo>|</mo>
<mover>
<mi>z</mi>
<mo>&RightArrow;</mo>
</mover>
<mo>,</mo>
<mover>
<mi>t</mi>
<mo>&RightArrow;</mo>
</mover>
<mo>,</mo>
<mi>&gamma;</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mi>D</mi>
<mi>i</mi>
<mi>r</mi>
<mi>i</mi>
<mi>c</mi>
<mi>h</mi>
<mi>l</mi>
<mi>e</mi>
<mi>t</mi>
<mrow>
<mo>(</mo>
<msub>
<mover>
<mi>&psi;</mi>
<mo>&RightArrow;</mo>
</mover>
<mi>l</mi>
</msub>
<mo>|</mo>
<msub>
<mover>
<mi>n</mi>
<mo>&RightArrow;</mo>
</mover>
<mi>l</mi>
</msub>
<mo>+</mo>
<mi>&gamma;</mi>
<mo>)</mo>
</mrow>
<mo>;</mo>
</mrow>
Wherein,Behavior pattern, program category, the quantity of behavior type for participating in calculating are represented respectively；Respectively program, interbehavior and behavior pattern set；Represent the set of behavior pattern in m-th of document, Φ
For program profile multinomialIn all program probability of happening Di Li Crays distribution, Ψ be interbehavior multinomial distributionIn
The Di Li Crays distribution of all interbehavior probability of happening, Θ is behavior pattern multinomial distributionIn all behavior patterns occur
The Di Li Crays distribution of probability；
Θ, Φ, Ψ are distributed based on Di Li Crays, and derive by chain rule the full probability of each behavior pattern：
Wherein,Respectively program, interbehavior and behavior pattern set, these three set inner elements wmn、tmn、
zmnRepresent respectively the program of the n-th viewing-data in m-th of document, the interbehavior of the n-th viewing-data in m-th of document, m-th
The behavior pattern of n-th viewing-data in document；Represent behavior pattern setMiddle removal zmnSet afterwards；
The full probability for being based ultimately upon Di Li Crays distribution Θ, Φ, Ψ and each behavior pattern solves the phase that Di Li Crays are distributed
Hope, be derived by equation below：
<mrow>
<msub>
<mi>&theta;</mi>
<mrow>
<mi>m</mi>
<mi>i</mi>
</mrow>
</msub>
<mo>=</mo>
<mfrac>
<mrow>
<msubsup>
<mi>n</mi>
<mi>m</mi>
<mi>i</mi>
</msubsup>
<mo>+</mo>
<mi>&alpha;</mi>
</mrow>
<mrow>
<msubsup>
<mi>&Sigma;</mi>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mrow>
<mi>K</mi>
<mo>&times;</mo>
<mi>L</mi>
</mrow>
</msubsup>
<mrow>
<mo>(</mo>
<msubsup>
<mi>n</mi>
<mrow>
<mi>m</mi>
<mo>+</mo>
<mi>&alpha;</mi>
</mrow>
<mi>i</mi>
</msubsup>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
<mo>;</mo>
</mrow>
<mrow>
<msub>
<mi>&phi;</mi>
<mrow>
<msub>
<mi>kV</mi>
<mi>w</mi>
</msub>
</mrow>
</msub>
<mo>=</mo>
<mfrac>
<mrow>
<msubsup>
<mi>n</mi>
<mi>k</mi>
<msub>
<mi>V</mi>
<mi>w</mi>
</msub>
</msubsup>
<mo>+</mo>
<mi>&beta;</mi>
</mrow>
<mrow>
<msubsup>
<mi>&Sigma;</mi>
<mrow>
<msub>
<mi>V</mi>
<mi>w</mi>
</msub>
<mo>=</mo>
<mn>1</mn>
</mrow>
<msub>
<mi>V</mi>
<mi>w</mi>
</msub>
</msubsup>
<mrow>
<mo>(</mo>
<msubsup>
<mi>n</mi>
<mi>k</mi>
<msub>
<mi>V</mi>
<mi>w</mi>
</msub>
</msubsup>
<mo>+</mo>
<mi>&beta;</mi>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
<mo>;</mo>
</mrow>
<mrow>
<msub>
<mi>&phi;</mi>
<mrow>
<msub>
<mi>lV</mi>
<mi>l</mi>
</msub>
</mrow>
</msub>
<mo>=</mo>
<mfrac>
<mrow>
<msubsup>
<mi>n</mi>
<mi>l</mi>
<msub>
<mi>V</mi>
<mi>l</mi>
</msub>
</msubsup>
<mo>+</mo>
<mi>&gamma;</mi>
</mrow>
<mrow>
<msubsup>
<mi>&Sigma;</mi>
<mrow>
<msub>
<mi>V</mi>
<mi>l</mi>
</msub>
<mo>=</mo>
<mn>1</mn>
</mrow>
<msub>
<mi>V</mi>
<mi>l</mi>
</msub>
</msubsup>
<mrow>
<mo>(</mo>
<msubsup>
<mi>n</mi>
<mi>l</mi>
<msub>
<mi>V</mi>
<mi>l</mi>
</msub>
</msubsup>
<mo>+</mo>
<mi>&gamma;</mi>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
<mo>;</mo>
</mrow>
Wherein,User is represented to V in kth kind rating typewThe viewing number of times of portion's program,Distribution is represented in l kind rows
For the V of typelThe number of times of interbehavior,Represent the frequency of i-th of behavior pattern in m-th of document；θmi、Respectively single behavior pattern, program, the probability of happening of interbehavior.
5. a kind of video resource Popularity prediction method according to claim 1, it is characterised in that the combination nerve net
Network model, the resource popularity of institute's statistics and behavior pattern matrix are inputted as neutral net, pre- by training generation
Model is surveyed, the video resource popularity in future is predicted according to the forecast model of training to be included：
The Nonlinear Mapping to behavior pattern matrix and the resource popularity of statistics, input layer are realized using BP neural network
Relevant with input/output argument with output layer neuron number, the rating type of the behavior pattern matrix program of input has K kinds, defeated
Go out for following video resource popularity, then input layer and output layer node number are respectively K and 1, and node in hidden layer is set to S
It is individual；
The behavior vector of the program of different rating types in the resource popularity of statistics, and behavior pattern matrix is read
Enter, then training data and test data are divided into by data are read in；
Start to initialize BP neural network, training method uses steepest descent method, and batch mode trains behavioral pattern data, then
Sample is inputted into BP neural network by the way of batch is trained, the error of each sample is calculated；Finally judge whether convergence, if
Do not restrain, weights are adjusted according to steepest descent method, until convergence, so as to obtain forecast model；
The test data of delimitation is inputted into BP neural network, the video resource popularity in future is predicted using forecast model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710409198.2A CN107222787A (en) | 2017-06-02 | 2017-06-02 | Video resource popularity prediction method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710409198.2A CN107222787A (en) | 2017-06-02 | 2017-06-02 | Video resource popularity prediction method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107222787A true CN107222787A (en) | 2017-09-29 |
Family
ID=59946918
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710409198.2A Pending CN107222787A (en) | 2017-06-02 | 2017-06-02 | Video resource popularity prediction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107222787A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107864405A (en) * | 2017-11-14 | 2018-03-30 | 国家新闻出版广电总局广播科学研究院 | A kind of Forecasting Methodology, device and the computer-readable medium of viewing behavior type |
| CN107909108A (en) * | 2017-11-15 | 2018-04-13 | 东南大学 | Edge cache system and method based on content popularit prediction |
| CN108259929A (en) * | 2017-12-22 | 2018-07-06 | 北京交通大学 | A kind of prediction of video active period pattern and caching method |
| CN109272165A (en) * | 2018-09-30 | 2019-01-25 | 江苏满运软件科技有限公司 | Register probability predictor method, device, storage medium and electronic equipment |
| CN109409921A (en) * | 2018-08-31 | 2019-03-01 | 阿里巴巴集团控股有限公司 | The recognition methods of the popularity of electronic equipment, device, server |
| CN109447246A (en) * | 2018-10-30 | 2019-03-08 | 北京字节跳动网络技术有限公司 | Method and apparatus for generating model |
| CN110139125A (en) * | 2019-06-18 | 2019-08-16 | 洛阳师范学院 | Video sharing method based on demand perception and caching resource under mobile radio network |
| CN110704614A (en) * | 2019-08-30 | 2020-01-17 | 中国平安人寿保险股份有限公司 | Information processing method and device for predicting user group type in application |
| US20210049687A1 (en) * | 2019-08-14 | 2021-02-18 | Royal Bank Of Canada | Systems and methods of generating resource allocation insights based on datasets |
| CN114785858A (en) * | 2022-06-20 | 2022-07-22 | 武汉格蓝若智能技术有限公司 | Resource active caching method and device applied to mutual inductor online monitoring system |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103209342A (en) * | 2013-04-01 | 2013-07-17 | 电子科技大学 | Collaborative filtering recommendation method introducing video popularity and user interest change |
| CN106028155A (en) * | 2016-06-22 | 2016-10-12 | 深圳大学 | Method and system for carrying out evolution analysis on popularity of network video |
| CN106570597A (en) * | 2016-11-14 | 2017-04-19 | 广州大学 | Content popularity prediction method based on depth learning under SDN architecture |
-
2017
- 2017-06-02 CN CN201710409198.2A patent/CN107222787A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103209342A (en) * | 2013-04-01 | 2013-07-17 | 电子科技大学 | Collaborative filtering recommendation method introducing video popularity and user interest change |
| CN106028155A (en) * | 2016-06-22 | 2016-10-12 | 深圳大学 | Method and system for carrying out evolution analysis on popularity of network video |
| CN106570597A (en) * | 2016-11-14 | 2017-04-19 | 广州大学 | Content popularity prediction method based on depth learning under SDN architecture |
Non-Patent Citations (3)
| Title |
|---|
| TINGYAO WU等: "On the Use of Reservoir Computing in Popularity Prediction", 《2010 2ND INTERNATIONAL CONFERENCE ON EVOLVING INTERNET》 * |
| YA ZHANG等: "A Time-Topic Coupled LDA Model for IPTV User Behaviors", 《 IEEE TRANSACTIONS ON BROADCASTING》 * |
| 何明一: "《神经计算 原理·语言·设计·应用》", 30 November 1992 * |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107864405A (en) * | 2017-11-14 | 2018-03-30 | 国家新闻出版广电总局广播科学研究院 | A kind of Forecasting Methodology, device and the computer-readable medium of viewing behavior type |
| CN107864405B (en) * | 2017-11-14 | 2020-12-01 | 国家新闻出版广电总局广播科学研究院 | Viewing behavior type prediction method, device and computer readable medium |
| CN107909108A (en) * | 2017-11-15 | 2018-04-13 | 东南大学 | Edge cache system and method based on content popularit prediction |
| CN107909108B (en) * | 2017-11-15 | 2021-06-11 | 东南大学 | Edge cache system and method based on content popularity prediction |
| CN108259929B (en) * | 2017-12-22 | 2020-03-06 | 北京交通大学 | Prediction and caching method for video active period mode |
| CN108259929A (en) * | 2017-12-22 | 2018-07-06 | 北京交通大学 | A kind of prediction of video active period pattern and caching method |
| CN109409921A (en) * | 2018-08-31 | 2019-03-01 | 阿里巴巴集团控股有限公司 | The recognition methods of the popularity of electronic equipment, device, server |
| CN109272165A (en) * | 2018-09-30 | 2019-01-25 | 江苏满运软件科技有限公司 | Register probability predictor method, device, storage medium and electronic equipment |
| CN109447246A (en) * | 2018-10-30 | 2019-03-08 | 北京字节跳动网络技术有限公司 | Method and apparatus for generating model |
| CN109447246B (en) * | 2018-10-30 | 2021-01-15 | 北京字节跳动网络技术有限公司 | Method and apparatus for generating a model |
| CN110139125A (en) * | 2019-06-18 | 2019-08-16 | 洛阳师范学院 | Video sharing method based on demand perception and caching resource under mobile radio network |
| CN110139125B (en) * | 2019-06-18 | 2021-01-22 | 洛阳师范学院 | Video sharing method based on demand perception and resource caching in wireless mobile network |
| US20210049687A1 (en) * | 2019-08-14 | 2021-02-18 | Royal Bank Of Canada | Systems and methods of generating resource allocation insights based on datasets |
| CN110704614A (en) * | 2019-08-30 | 2020-01-17 | 中国平安人寿保险股份有限公司 | Information processing method and device for predicting user group type in application |
| CN110704614B (en) * | 2019-08-30 | 2023-09-19 | 中国平安人寿保险股份有限公司 | Information processing method and device for predicting user group type in application |
| CN114785858A (en) * | 2022-06-20 | 2022-07-22 | 武汉格蓝若智能技术有限公司 | Resource active caching method and device applied to mutual inductor online monitoring system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107222787A (en) | Video resource popularity prediction method | |
| CN103793537B (en) | System for recommending individual music based on multi-dimensional time series analysis and achieving method of system | |
| CN109902708A (en) | A kind of recommended models training method and relevant apparatus | |
| CN106339507B (en) | Streaming Media information push method and device | |
| CN105912729A (en) | Three-dimensional time sequence dynamic model based mobile phone game recommendation system | |
| CN106844530A (en) | Training method and device of a kind of question and answer to disaggregated model | |
| CN106294830A (en) | The recommendation method and device of multimedia resource | |
| CN103198086A (en) | Information processing device, information processing method, and program | |
| CN110474944B (en) | Network information processing method, device and storage medium | |
| US20220172260A1 (en) | Method, apparatus, storage medium, and device for generating user profile | |
| Wang et al. | Recurrent recommendation with local coherence | |
| CN110457283A (en) | A kind of teaching comment based on data mining and recommender system and its method | |
| CN109902823A (en) | A kind of model training method and equipment based on generation confrontation network | |
| CN109460503A (en) | Answer input method, device, storage medium and electronic equipment | |
| CN116629937A (en) | Marketing strategy recommendation method and device | |
| CN114218488A (en) | Information recommendation method and device based on multi-modal feature fusion and processor | |
| CN108647364A (en) | A kind of prediction recommendation method based on mobile terminal application data | |
| Mussumeci et al. | Reconstructing news spread networks and studying its dynamics | |
| CN109993636A (en) | A kind of interpretable recommendation score prediction technique of exercise of fusion aspect information | |
| CN113424204A (en) | Method for performing legal license checks of digital content | |
| Tuan et al. | Knowledge injection into dialogue generation via language models | |
| CN115034847A (en) | Product recommendation method, system, storage medium and equipment based on deep learning | |
| Fan et al. | Context-aware ubiquitous web services recommendation based on user location update | |
| CN113761272A (en) | Data processing method, data processing equipment and computer readable storage medium | |
| Wang | Forecast model of TV show rating based on convolutional neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170929 |
|
| RJ01 | Rejection of invention patent application after publication |