CN109413149A - Information distribution control method, system, server and computer readable storage medium - Google Patents
Information distribution control method, system, server and computer readable storage medium Download PDFInfo
- Publication number
- CN109413149A CN109413149A CN201811097810.8A CN201811097810A CN109413149A CN 109413149 A CN109413149 A CN 109413149A CN 201811097810 A CN201811097810 A CN 201811097810A CN 109413149 A CN109413149 A CN 109413149A
- Authority
- CN
- China
- Prior art keywords
- information
- distribution
- cluster
- dimensionality reduction
- feature
- 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
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L67/00—Network arrangements or protocols for supporting network services or applications
- H04L67/50—Network services
- H04L67/55—Push-based network services
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/231—Hierarchical techniques, i.e. dividing or merging pattern sets so as to obtain a dendrogram
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L67/00—Network arrangements or protocols for supporting network services or applications
- H04L67/01—Protocols
- H04L67/10—Protocols in which an application is distributed across nodes in the network
- H04L67/1095—Replication or mirroring of data, e.g. scheduling or transport for data synchronisation between network nodes
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Information Transfer Between Computers (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The present invention provides a kind of information distribution control method, system, server and computer readable storage mediums.Information distribution control method, comprising the following steps: S100: the access data of information in a historical time are obtained, to form a historical data；S200: to historical data dimensionality reduction, the matrix for obtaining a cluster feature with information is forecast sample；S300: obtaining the current data of information, calculates the click probability of each information；S400: calculating forecast sample and clicks the error of probability, parameter when reversed amendment is to historical data dimensionality reduction, to obtain an expectation parameter；S500: based on desired parameter to historical data dimensionality reduction, the matrix for obtaining a cluster feature with information is distribution sample；S600: to distribution sample sequence, information and the distribution being located in the cluster of preceding default item are extracted.After adopting the above technical scheme, the higher information of user's request rate preferentially can be distributed in the server in each region, to improve the utilization rate to information distribution mechanisms.
Description
Technical field
The present invention relates to Informationflow Control field more particularly to a kind of information distribution control method, system, server and meters
Calculation machine readable storage medium storing program for executing.
Background technique
With the fast development of intelligent terminal, user carries out miscellaneous operation using intelligent terminal and receives various each
The information of sample.For example, getting the Streaming Medias such as audio, video by the intelligent terminal networked and listening to and watch；By
The intelligent terminal of networking receives text class message etc..To send above- mentioned information and message to user, it is provided with the master of sending function
Body is usually the application APP being mounted in intelligent terminal.The operation manufacturer of these application APPs is to improve user to receive
Information often can be made to distribute by the agility of information.
So-called information distribution, such as video distribution, the word message syntax, audio distribution, for operation, manufacturer asks user
Ask the information of acquisition in addition to self-built director server, the server that will also store and be issued on each region, then user is by answering
When obtaining a certain message with PROGRAMMED REQUESTS, receives the operation at the request and issuing message to user and just no longer need to by director server
It completes, if the message of request has stored in the server on each region, these servers will directly be sent to user.
In such a way that information is distributed, the bandwidth occupancy to director server is reduced, after this system load is dispersed, significantly reduces fortune
Seek the operation pressure of manufacturer.
In existing technology, when information is distributed in the content of selection distribution, the mode randomly selected is usually taken, it can not
According to preset distribution quantity to distribute, the information being distributed in the server in each region is easily led to, is requested by a user
Frequency is not high, and is requested by a user the higher information of frequency, in the not distributed period of service to each region, to can not have
Effect utilizes the server distributed.And information is distributed as a ring important in the Content Delivery Network (CDN) based on Video service,
The judgement usually made in distribution is that can only obtain yes/no result, can not effectively control the total amount of distribution.A but net
The amount of video that can be stored in network data center (IDC) is limited, if being unable to control super-distribution amount, distribution effect will give a discount greatly
Button.
Therefore, it is necessary to a kind of novel information distribution control methods, and intelligently judgement is distributed in each region server
Information has maximally utilized information distribution mechanisms.
Summary of the invention
In order to overcome the above technical defects, the purpose of the present invention is to provide a kind of information distribution control methods, system, clothes
Business device and computer readable storage medium, the higher information of user's request rate is preferentially distributed in the server in each region, with
Improve the utilization rate to information distribution mechanisms.
The invention discloses a kind of information distribution control methods, which comprises the following steps:
S100: the access data of information in a historical time are obtained, to form a historical data；
S200: to the historical data dimensionality reduction, the matrix for obtaining a cluster feature with the information is forecast sample；
S300: obtaining the current data of the information, calculates the click probability of each information；
S400: the forecast sample and the error for clicking probability are calculated, reversed amendment is to the historical data dimensionality reduction
When parameter, with obtain one expectation parameter；
S500: based on the expectation parameter to the historical data dimensionality reduction, a cluster feature with the information is obtained
Matrix be distribution sample；
S600: sorting to the distribution sample, extracts information and the distribution being located in the cluster of preceding default item.
Preferably, the step S100 includes:
S110: obtaining the access data of n video information in a historical time t, wherein access number evidence includes by institute
It states n video information and is divided into m cluster feature, to form the historical data, wherein t, n, m are positive integer.
Preferably, the step S200 includes:
S210: it for m cluster feature of the access data of the video information, is based on:
Ln=tanh (wn·Ln-1+bn)
Layer-by-layer dimensionality reduction to a 1* video information with characteristic range cluster feature matrix, wherein wnFor weight, bnFor
Deviation, tanh () are activation primitive, and the characteristic range is (- 1,1).
Preferably, the step S210 includes:
S211: with W0It is dimension that=m cluster feature, which is * 1024, is based on L1=tanh (L0·w0+b0) calculate the first dimensionality reduction
Feature L1；
S212: with W1=256*1024 is dimension, is based on L2=tanh (L1·w1+b1) calculate the second dimensionality reduction feature L2；
S213: with W2=256* is dimension to the access number of users of the video information, is based on L3=tanh (L2·w2+b2)
Calculate third dimensionality reduction feature L3；
S214: with W3=access number of users * access number of users is dimension, is based on L4=sigmoid (L3·w3+b3) calculate 1*
The matrix of cluster feature；
Wherein
Preferably, the step S300 includes:
S310: obtaining the current data of the information, wherein the current data packet is included to belonging to same cluster feature
The click volume of information and click total amount to all information；
S320: calculating the click volume and the ratio for clicking total amount, form the click probability to each cluster feature,
It is wherein described to click the matrix that probability is 1* cluster feature.
Preferably, the step S400 includes:
S410: the forecast sample and the mean square error for clicking probability are calculated；
S420: it is based on
Wn=Wn-γn·ΔLn
Corrected parameter wn, wherein Δ Ln=Δ Ln+1/ΔWn=(Δ Ln+1/Δtanh)*(Δtanh/ΔWn),
Or Δ Ln=Δ Ln+1/ΔWn=(Δ Ln+1/Δsigmoid)*(Δsigmoid/ΔWn)；
S430: step S420 described in iteration, until the mean square error is less than in an anticipation error；
S440: current w is extractednIt is expected parameter.
Preferably, the step S600 includes:
S610: the distribution sample descending is arranged；
S620: information and the distribution being located in the cluster of preceding default item are extracted.
The invention also discloses a kind of servers, including processor and storage equipment, the storage equipment to be stored with calculating
Machine program, the processor call and realize information distribution control method as described above when executing the computer program.
The invention also discloses a kind of information to distribute control system, comprising:
Module is obtained, the access data of information in a historical time are obtained, to form a historical data；
Processing module receives the historical data, to the historical data dimensionality reduction, obtains a cluster with the information
The matrix of feature is forecast sample；
Computing module obtains the current data of the information, calculates the click probability of each information, and calculates the prediction
Sample and the error for clicking probability, parameter when reversed amendment is to the historical data dimensionality reduction, to obtain an expectation parameter；
The processing module receives the expectation parameter, and is obtained based on the expectation parameter to the historical data dimensionality reduction
The matrix for obtaining a cluster feature with the information is distribution sample；
Distribution module sorts to the distribution sample, extracts information and the distribution being located in the cluster of preceding default item.
The present invention discloses a kind of computer readable storage medium again, is stored thereon with computer program, the computer
Information distribution control method as described above is realized when program is executed by processor.
After above-mentioned technical proposal, compared with prior art, have the advantages that
1. accurately understanding the access preference of user based on the prediction to historical data, more targetedly being pushed away to user
Send relevant content；
2. also selecting control distribution total amount distribution content, so as to belong to user more inclined for the video of distribution
Good content, greatly reduces the load of primary server.
Detailed description of the invention
Fig. 1 is the flow diagram for meeting information distribution control method in one embodiment of the present invention；
Fig. 2 is the flow diagram for meeting information distribution control method in a further preferred embodiments of the invention；
Fig. 3 is the schematic diagram for meeting dimensionality reduction calculating process in one embodiment of the present invention；
Fig. 4 is to meet the sequence schematic diagram for distributing sample in one embodiment of the present invention；
Fig. 5 is the structural schematic diagram for meeting information distribution control system in one embodiment of the present invention.
Specific embodiment
Below in conjunction with attached drawing, the advantages of the present invention are further explained with 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.
It is only to be not intended to be limiting the disclosure merely for for the purpose of describing particular embodiments in the term that the disclosure uses.
The "an" of the singular used in disclosure and the accompanying claims book, " described " and "the" are also intended to including majority
Form, unless the context clearly indicates other meaning.It is also understood that term "and/or" used herein refers to and wraps
It may be combined containing one or more associated any or all of project listed.
It will be appreciated that though various information, but this may be described using term first, second, third, etc. in the disclosure
A little information should not necessarily be limited by these terms.These terms are only used to for same type of information being distinguished from each other out.For example, not departing from
In the case where disclosure range, the first information can also be referred to as the second information, and similarly, the second information can also be referred to as
One information.Depending on context, word as used in this " if " can be construed to " ... when " or " when ...
When " or " in response to determination "
In the description of the present invention, it is to be understood that, term " longitudinal direction ", " transverse direction ", "upper", "lower", "front", "rear",
The orientation or positional relationship of the instructions such as "left", "right", "vertical", "horizontal", "top", "bottom" "inner", "outside" is based on attached drawing institute
The orientation or positional relationship shown, is merely for convenience of description of the present invention and simplification of the description, rather than the dress of indication or suggestion meaning
It sets or element must have a particular orientation, be constructed and operated in a specific orientation, therefore should not be understood as to limit of the invention
System.
In the description of the present invention, unless otherwise specified and limited, it should be noted that term " installation ", " connected ",
" connection " shall be understood in a broad sense, for example, it may be mechanical connection or electrical connection, the connection being also possible to inside two elements can
, can also indirectly connected through an intermediary, for the ordinary skill in the art to be to be connected directly, it can basis
Concrete condition understands the concrete meaning of above-mentioned term.
In subsequent description, it is only using the suffix for indicating such as " module ", " component " or " unit " of element
Be conducive to explanation of the invention, itself there is no specific meanings.Therefore, " module " can mixedly make with " component "
With.
Refering to fig. 1, control is distributed to information such as such as video, audio, texts to meet in one embodiment of the present invention
The method flow schematic diagram of system.In this embodiment, it is screened for the content to distribution information, it need to be first for user to information
Preference work understands, with qualitative and directionally do the information content of suitable user preference and distribute.Specifically, information distributes controlling party
Method, comprising the following steps:
S100: the access data of information in a historical time are obtained, to form a historical data；
The information of all information or all possible distributions is obtained, the mode of acquisition can out of database gathering information
Itself or the related label with information characteristics value, such as the classification (cluster labels) of the information, the size of the information, the information
Access time, access number of the information etc..When acquisition, available current time is the benchmark time, retrodicts one forward and goes through
The history time, 1 day, 5 days, 10 days, 15 days such as unit of day, or 10 hours, 20 hours, 50 hours as unit of hour
Deng.In the information in above-mentioned historical time or with after information-related characteristic value acquisition, it is classified as the historical data of the information.
By the interception to historical data, it can help to run manufacturer and recognize that user related with bulk information accesses rule
Rule, as information user's access in which cluster is more frequent, which period when user it is more to the data volume of message reference, respectively
User preference difference in area etc., to learn and predict the hobby of entire user's sample from history.
S200: to historical data dimensionality reduction, the matrix for obtaining a cluster feature with information is forecast sample；
The content as included in historical data is more, and the data volume of information itself is also larger, to simplify to history
The treatment process of data, will be to historical data dimensionality reduction, until the historical information that the historical data after dimensionality reduction is reflected is and cluster
The relevant feature vector of feature.Heretofore described cluster feature is user, application program operator or information itself to letter
The definition of the type of breath, such as video class, audio class, also or the love class of secondary classification, animation class, record class etc..And
It is obtained after dimensionality reduction in the embodiment, it is the matrix of a cluster feature, such as when 100 clusters of all Information commons,
Then cluster feature is 100, after dimensionality reduction it is obtained also be 1*100 matrix, as subsequent prediction and compare it is pre-
Test sample sheet.
It is understood that an i.e. information may have multiple clusters since an information may have multiple labels,
It, can not be according to for the relationship of information and cluster feature that is, an information possibly is present in different cluster features
Stringent one-to-one or one-to-many relationship.
S300: obtaining the current data of information, calculates the click probability of each information；
After to historical data reorganization and analysis, access data of the user under actual conditions to information will be dealt with again and
The standard of verifying.Such as the request information under current time or on the day of user to the amount of access of information, when to all information
Data collection after the completion of, the click probability to each information is calculated.The click probability reflects under all information,
User clicks the size of a possibility that a certain information.Such as click volume or activation number of a certain information in some day are 10,000 times,
And click volume or activation number of the full information in this day are 100,000 times, then can be shown that user is to the click probability of the information
10%, every 10 times the activation of all information or point are hit, they will be once the click or activation to the information.
By clicking the combination of probability and cluster feature, it can equally reflect which cluster belongs to user's click possibility more
High cluster, which information belong to user's more preference again, click higher information of possibility etc..
S400: calculating forecast sample and clicks the error of probability, parameter when reversed amendment is to historical data dimensionality reduction, to obtain
Obtain an expectation parameter；
Due in step S200 and S300, respectively to the research of historical data and to the analysis of current data, but nothing
Method reflects under future state that the possible preference variation of user and current data do not have generality, can only reflect a certain moment
Under user's phenomenon.Therefore, in step S400, forecast sample will be calculated and clicks the error of probability, i.e., forecast sample is in shape
At in the process, the difference of situation is accessed with practical emerging user.It is pre- by the calculating and constantly tune ginseng modification of error
The forming process of test sample sheet, i.e., parameter when reversed amendment is to historical data dimensionality reduction, ultimately forms one closer to reality
Dimension-reduction algorithm.That is, the click that forecast sample was more close under current time is general during error is constantly reduced
Rate, forecast sample can be used as to following judgement, and the matrix of output also more meets user in the following possible access preference
Change with preference.For example, when in a certain period, the access hot spot of user changes, then according to current number in step S300
According to the click probability being calculated, will also change, it therefore, can be by repeatedly or regularly executing step S400, constantly
Ground corrected parameter obtains the expectation parameter under different moments, comes so that updated reduction process is more accurate, prediction result
More fitting actual capabilities there is a situation where.
S500: based on desired parameter to historical data dimensionality reduction, the matrix for obtaining a cluster feature with the information is
Distribute sample；
After acquisition has desired parameter, it would be desirable to which parameter is inserted in the calculating process to historical data dimensionality reduction, and again
It executes and is operated with dimensionality reduction identical in step S200, it is obtained at this time, for fit in the expectation parameter of actual conditions as base
The resulting matrix of plinth operation.Due to identical as the dimensionality reduction operation calculating process in step S200, matrix obtained in step S500
It is similarly the matrix of 1* cluster feature, but it is understood that, it is obtained at this time poly- due to relying on the expectation parameter of standard
Category feature is the characteristic value for mapping the access preference of user under current state, and size directly reflects user to these information
Situation, the i.e. temperature of information are accessed, therefore, which can be used as making the distribution sample controlled to video distribution.
S600: sorting to the distribution sample, extracts information and the distribution being located in the cluster of preceding default item.
Finally using distributing sample and being ranked up, such as it is ranked up according to information content in clustering, according to cluster feature
Be ranked up, and to the default K of sequence setting one after sequence, after arranging in the sequence in preceding default item K to cluster
Interior information is extracted, then these information extracted are that user accesses more frequent information, is distributed to respectively to these information
At the server in region, then information distribution mechanisms are utilized to maximizing, and differentiates that user accesses preference.When being distributed to each region
Server in information called more frequent when, then the called number of primary server is just fewer, then primary server has
Bandwidth up and down can reserve as monitoring, receive the functions such as new information so that a whole set of communication system operating is more healthy.
Referring to Fig.2, to meet in information distribution control method and each step in a further preferred embodiments of the invention
To historical data, current data and the calculation for it is expected parameter, wherein the information in the embodiment is that user visits video class
The video information asked.Specifically:
Step S100 includes: S110: the access data of n video information in a historical time t is obtained, wherein accessing data
Including n video information is divided into m cluster feature, with history of forming data, wherein t, n, m are positive integer.
In step s 110, to the acquisition of historical data, historical time t, video need to be believed by the operator of application program
The cluster feature quantity m work that the number n of breath and the n video information are divided or mark is arranged.Naturally, to guarantee history number
According to universality, the period on long period axis may be selected in historical time t, and the stored institute of director server may be selected in video information
There is video information, the number having is the n.And the number for all cluster features that these video informations are included into is m, then
It is understood that the original dimensions of historical data are t*n*m.Above parameter t, n, m are due to being the number under actual condition
Information, therefore be positive integer.
Step S200 includes: S210: for m cluster feature of the access data of video information, being based on: Ln=tanh
(wn·Ln-1+bn) successively dimensionality reduction to a 1* video information with characteristic range cluster feature matrix, wherein wnFor weight,
bnFor deviation, tanh () is activation primitive, and characteristic range is (- 1,1).
Refering to Fig. 3, for calculate the 1* video information cluster feature matrix, activation primitive tanh () will be utilized, with wn
For weight, bnFor deviation, the data volume of compression histories data by way of dropping power.And the forecast sample due to being ultimately formed
Required to each feature vector compared with clicking probability, therefore in the matrix obtained after dimensionality reduction, the section being in (- 1,1)
It is interior, it is implemented as follows:
S211: with W0It is dimension that=m cluster feature, which is * 1024, and with the W0For weight parameter, it is based on L1=tanh
(L0·w0+b0) calculate the first dimensionality reduction feature L1；
S212: again, another weight parameter W is selected1, while with W1=256*1024 is dimension, is based on L2=tanh
(L1·w1+b1) calculate the second dimensionality reduction feature L2；
S213: during third time dimensionality reduction calculates, with W2=256* is dimension to the access number of users of the video information, is based on
L3=tanh (L2·w2+b2) calculate third dimensionality reduction feature L3；
S214: finally, with W3=access number of users * access number of users is dimension, is based on L4=sigmoid (L3·w3+b3)
Calculate the matrix of 1* cluster feature；Wherein
The matrix of last obtained 1* cluster feature, the cluster feature of every a line are represented in selected all videos
In information, the video in each cluster accounts for the number that all video informations are clicked access by the number that user clicks access
Ratio, when such as a cluster feature is 0.1, then it represents that the number that the video information for being included in the cluster feature is clicked access accounts for
All videos are clicked the 10% of access, that is to say, that just have 1 user's click in every 10 users and belong to the cluster feature
Interior video information is to access.
Further, step S300 includes:
S310: obtaining the current data of information, and wherein current data includes the point to the information for belonging to same cluster feature
The amount of hitting and click total amount to all information；
The weight parameter as employed in above-mentioned steps S210 and straggling parameter are rule of thumb to choose, be
Weight parameter therein and straggling parameter are modified, the practical click feelings to video information of user under current state will be chosen
Condition.Specifically, the current data that video information need to be obtained, if which video information is clicked and was accessed by user, these videos
Information belongs to which cluster feature and user and accesses total amount to the click of all video informations.Above-mentioned current data is shown
Obtain the actual access situation of user in the same day or certain time period.
S320: calculating click volume and clicks the ratio of total amount, the click probability to each cluster feature is formed, wherein clicking
Probability is the matrix of 1* cluster feature.
Based on above-mentioned current data, calculates and the click volume of the video information in a certain cluster feature is believed with to all videos
The ratio of the click total amount of breath, the ratio reflect the video information that belongs in same cluster feature by the click temperature of user,
Therefore, which reflects the click probability of each cluster feature.Finally, the click probabilistic by each cluster feature,
It forms one and clicks the matrix that probability is 1* cluster feature.
It is understood that obtained two matrix of step S200 and S300, respectively and is based on prediction algorithm and is based on
The resulting data of actual conditions must have difference between two data, and therefore, step S400 includes:
S410: calculating forecast sample and clicks the mean square error of probability；
For example, extract same cluster feature in forecast sample and click the numerical value in probability, after the acquisition of all numerical value,
It is based on
It calculates forecast sample and clicks the mean square error of cluster feature in probability.It is understood that the mean square error is got over
Greatly, indicate that the selection of weight parameter and straggling parameter in forecast sample does not meet actual conditions more, vice versa.
S420: in view of forecast sample and clicking the difference between probability, click probability for forecast sample to be increasingly bonded, will
It is based on
Wn=Wn-γn·ΔLn
Corrected parameter wn, wherein Δ Ln=Δ Ln+1/ΔWn=(Δ Ln+1/Δtanh)*(Δtanh/ΔWn) or Δ Ln=
ΔLn+1/ΔWn=(Δ Ln+1/Δsigmoid)*(Δsigmoid/ΔWn).For example, by taking above-described embodiment as an example, weight parameter
W0、W1、W2、W3It need to successively correct, by reference when amendment, wherein Δ LtotalTo click total amount:
W3=W3-γ3·ΔL3, Δ L3=Δ Ltotal/ΔW3=(Δ Ltotal/Δsigmoid)*(Δsigmoid/Δ
W3)；
W2=W2-γ2·ΔL2, Δ L2=Δ L3/ΔW2=(Δ L3/Δtanh)*(Δtanh/ΔW2)；
W1=W1-γ1·ΔL1, Δ L1=Δ L2/ΔW1=(Δ L2/Δtanh)*(Δtanh/ΔW1)；
W0=W0-γ0·ΔL0, Δ L0=Δ L1/ΔW0=(Δ L1/Δtanh)*(Δtanh/ΔW0)；
Wherein γnFor the above-mentioned learning efficiency based on Gradient learning method corrected parameter.
S430: iterative step S420, until mean square error is less than in an anticipation error；
Constantly repeat the above steps S420, until the mean square error between forecast sample and click probability is less than an anticipation error
In e, which can be e < 1.0-37, then can be considered that the mean square error is in tolerance interval.
S440: current w is extractednIt is expected parameter.
By the deep learning of step S410-S430, lift make under current state mean square error be less than anticipation error when
Under wnIt is expected parameter.
Refering to Fig. 4, it is based on desired parameter, replaces original parameter W0、W1、W2And W3, recalculate forecast sample.Due to wherein
Parameter correct according to the actual situation, therefore it is point that the forecast sample after recalculating abides by the data actually occurred completely
Send out sample.Executing step S600 afterwards includes:
S610: to distribution sample descending arrangement；
According to the numerical values recited of cluster feature, to distribution sample descending arrangement.It is got over that is, being arranged in distribution sample
The probability that high cluster feature is clicked by user is higher.
S620: information and the distribution being located in the cluster of preceding default item are extracted
Finally, the distribution to video information is no longer distributed at random, but extract the letter in the cluster of preceding default item
Breath is to distribute.For example, default item set by the operator of application program is 500, then being located at preceding 500 video informations will
It is extracted and is distributed at secondary server.Video information if the setting of default item is so that extract, in a certain cluster feature
When can not all be extracted, for example, default item is still 500, the video information in the cluster feature of arrangement first is belonged to
It is 300, the video information belonged in the cluster feature of arrangement second is 300, then preceding 300 video informations will be divided
Hair, and remaining 200 planned numbers that can distribute will will extract at random from the video information in the cluster feature of arrangement second, or according to
It clicks probability to arrange again in same cluster feature, extracts first 200.
The invention also discloses a kind of servers, including processor and storage equipment, storage equipment to be stored with computer journey
Sequence, processor call and realize information distribution control method described in any embodiment as above when executing computer program.
Refering to Fig. 5, the invention also discloses a kind of information to distribute control system, comprising: module is obtained, when obtaining a history
The access data of interior information, to form a historical data；Processing module connect with module is obtained, reception historical data, to going through
History Data Dimensionality Reduction, the matrix for obtaining a cluster feature with information is forecast sample；Computing module is connect with processing module,
The current data for obtaining information, calculates the click probability of each information, and calculates forecast sample and click the error of probability, reversely
Parameter when correcting to historical data dimensionality reduction, to obtain an expectation parameter；After desired parameter is calculated, computing module is by the phase
Hope parameter feedback to processing module, processing module receives expectation parameter, and obtains one to historical data dimensionality reduction based on desired parameter
The matrix of cluster feature with information is distribution sample；Distribution module extracts to distribution sample sequence and is located at preceding default item
Information and distribution in cluster.
The present invention discloses a kind of computer readable storage medium again, is stored thereon with computer program, computer program
Information distribution control method described in any embodiment as above is realized when being executed by processor.Computer readable storage medium can be
The operation such as intelligent terminal, server, repeater.
Intelligent terminal can be implemented in a variety of manners.For example, terminal described in the present invention may include such as moving
Phone, smart phone, laptop, PDA (personal digital assistant), PAD (tablet computer), PMP (put by portable multimedia broadcasting
Device), the fixed terminal of the intelligent terminal of navigation device etc. and such as number TV, desktop computer etc..Hereinafter it is assumed that eventually
End is intelligent terminal.However, it will be understood by those skilled in the art that other than the element for being used in particular for mobile purpose, root
It can also apply to the terminal of fixed type according to the construction of embodiments of the present invention.
It should be noted that the embodiment of the present invention has preferable implementation, and not the present invention is made any type of
Limitation, any one skilled in the art change or are modified to possibly also with the technology contents of the disclosure above equivalent effective
Embodiment, as long as without departing from the content of technical solution of the present invention, it is to the above embodiments according to the technical essence of the invention
Any modification or equivalent variations and modification, all of which are still within the scope of the technical scheme of the invention.
Claims (10)
1. a kind of information distribution control method, which comprises the following steps:
S100: the access data of information in a historical time are obtained, to form a historical data；
S200: to the historical data dimensionality reduction, the matrix for obtaining a cluster feature with the information is forecast sample；
S300: obtaining the current data of the information, calculates the click probability of each information；
S400: calculating the forecast sample and the error for clicking probability, when reversed amendment is to the historical data dimensionality reduction
Parameter, to obtain an expectation parameter；
S500: based on the expectation parameter to the historical data dimensionality reduction, the square of a cluster feature with the information is obtained
Battle array is distribution sample；
S600: sorting to the distribution sample, extracts information and the distribution being located in the cluster of preceding default item.
2. information distribution control method as described in claim 1, which is characterized in that
The step S100 includes:
S110: obtaining the access data of n video information in a historical time t, wherein access number evidence includes by the n
Video information is divided into m cluster feature, and to form the historical data, wherein t, n, m are positive integer.
3. information distribution control method as claimed in claim 2, which is characterized in that
The step S200 includes:
S210: it for m cluster feature of the access data of the video information, is based on:
Ln=tanh (wn·Ln-1+bn)
Layer-by-layer dimensionality reduction to a 1* video information with characteristic range cluster feature matrix, wherein wnFor weight, bnIt is inclined
Difference, tanh () are activation primitive, and the characteristic range is (- 1,1).
4. information distribution control method as claimed in claim 3, which is characterized in that
The step S210 includes:
S211: with W0It is dimension that=m cluster feature, which is * 1024, is based on L1=tanh (L0·w0+b0) calculate the first dimensionality reduction feature
L1；
S212: with W1=256*1024 is dimension, is based on L2=tanh (L1·w1+b1) calculate the second dimensionality reduction feature L2；
S213: with W2=256* is dimension to the access number of users of the video information, is based on L3=tanh (L2·w2+b2) calculate
Third dimensionality reduction feature L3；
S214: with W3=access number of users * access number of users is dimension, is based on L4=sigmoid (L3·w3+b3) calculate 1* cluster
The matrix of feature；
Wherein
5. information distribution control method as described in claim 1, which is characterized in that
The step S300 includes:
S310: obtaining the current data of the information, wherein the current data packet is included to the information for belonging to same cluster feature
Click volume and click total amount to all information；
S320: calculating the click volume and the ratio for clicking total amount, form the click probability to each cluster feature, wherein
It is described to click the matrix that probability is 1* cluster feature.
6. information distribution control method as claimed in claim 2, which is characterized in that
The step S400 includes:
S410: the forecast sample and the mean square error for clicking probability are calculated；
S420: it is based on
Wn=Wn-γn·ΔLn
Corrected parameter wn, wherein Δ Ln=Δ Ln+1/ΔWn=(Δ Ln+1/Δtanh)*(Δtanh/ΔWn),
Or Δ Ln=Δ Ln+1/ΔWn=(Δ Ln+1/Δsigmoid)*(Δsigmoid/ΔWn)；
S430: step S420 described in iteration, until the mean square error is less than in an anticipation error；
S440: current w is extractednIt is expected parameter.
7. information distribution control method as described in claim 1, which is characterized in that
The step S600 includes:
S610: the distribution sample descending is arranged；
S620: information and the distribution being located in the cluster of preceding default item are extracted.
8. a kind of server, including processor and storage equipment, the storage equipment are stored with computer program, feature exists
In the processor is called and realized when executing the computer program such as the described in any item information distributions of claim 1-7
Control method.
9. a kind of information distributes control system characterized by comprising
Module is obtained, the access data of information in a historical time are obtained, to form a historical data；
Processing module receives the historical data, to the historical data dimensionality reduction, obtains a cluster feature with the information
Matrix be forecast sample；
Computing module obtains the current data of the information, calculates the click probability of each information, and calculates the forecast sample
With the error for clicking probability, parameter when reversed amendment is to the historical data dimensionality reduction, to obtain an expectation parameter；
The processing module receives the expectation parameter, and obtains one to the historical data dimensionality reduction based on the expectation parameter
The matrix of cluster feature with the information is distribution sample；
Distribution module sorts to the distribution sample, extracts information and the distribution being located in the cluster of preceding default item.
10. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the computer program
Such as claim 1-7 described in any item information distribution control methods are realized when being executed by processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811097810.8A CN109413149B (en) | 2018-09-19 | 2018-09-19 | Information distribution control method, system, server, and computer-readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811097810.8A CN109413149B (en) | 2018-09-19 | 2018-09-19 | Information distribution control method, system, server, and computer-readable storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109413149A true CN109413149A (en) | 2019-03-01 |
| CN109413149B CN109413149B (en) | 2021-05-18 |
Family
ID=65464990
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811097810.8A Active CN109413149B (en) | 2018-09-19 | 2018-09-19 | Information distribution control method, system, server, and computer-readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109413149B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112402978A (en) * | 2020-11-13 | 2021-02-26 | 上海幻电信息科技有限公司 | Map generation method and device |
| CN114666655A (en) * | 2022-03-15 | 2022-06-24 | 北京达佳互联信息技术有限公司 | Video processing method and device, electronic equipment and storage medium |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102761572A (en) * | 2011-04-28 | 2012-10-31 | 腾讯科技（深圳）有限公司 | Method and system for dynamically adjusting data flow |
| US20140324532A1 (en) * | 2013-04-24 | 2014-10-30 | International Business Machines Corporation | System and method for modeling and forecasting cyclical demand systems with dynamic controls and dynamic incentives |
| CN104200290A (en) * | 2014-09-26 | 2014-12-10 | 国家电网公司 | Wind power forecast method |
| CN105407004A (en) * | 2015-12-08 | 2016-03-16 | 清华大学深圳研究生院 | Method and device for performing content distribution based on edge wireless hotspots |
| CN105871734A (en) * | 2015-01-22 | 2016-08-17 | 阿里巴巴集团控股有限公司 | Method and device for acquiring traffic distribution information of local domain name system |
| CN106201828A (en) * | 2016-07-18 | 2016-12-07 | 云南电网有限责任公司信息中心 | A kind of virtual-machine fail detection method based on data mining and system |
| CN107734482A (en) * | 2017-09-11 | 2018-02-23 | 西安电子科技大学 | The content distribution method unloaded based on D2D and business |
| CN108629630A (en) * | 2018-05-08 | 2018-10-09 | 广州太平洋电脑信息咨询有限公司 | A kind of feature based intersects the advertisement recommendation method of joint deep neural network |
| CN109102101A (en) * | 2017-06-21 | 2018-12-28 | 北京金风科创风电设备有限公司 | Wind speed prediction method and system for wind power plant |
-
2018
- 2018-09-19 CN CN201811097810.8A patent/CN109413149B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102761572A (en) * | 2011-04-28 | 2012-10-31 | 腾讯科技（深圳）有限公司 | Method and system for dynamically adjusting data flow |
| US20140324532A1 (en) * | 2013-04-24 | 2014-10-30 | International Business Machines Corporation | System and method for modeling and forecasting cyclical demand systems with dynamic controls and dynamic incentives |
| CN104200290A (en) * | 2014-09-26 | 2014-12-10 | 国家电网公司 | Wind power forecast method |
| CN105871734A (en) * | 2015-01-22 | 2016-08-17 | 阿里巴巴集团控股有限公司 | Method and device for acquiring traffic distribution information of local domain name system |
| CN105407004A (en) * | 2015-12-08 | 2016-03-16 | 清华大学深圳研究生院 | Method and device for performing content distribution based on edge wireless hotspots |
| CN106201828A (en) * | 2016-07-18 | 2016-12-07 | 云南电网有限责任公司信息中心 | A kind of virtual-machine fail detection method based on data mining and system |
| CN109102101A (en) * | 2017-06-21 | 2018-12-28 | 北京金风科创风电设备有限公司 | Wind speed prediction method and system for wind power plant |
| CN107734482A (en) * | 2017-09-11 | 2018-02-23 | 西安电子科技大学 | The content distribution method unloaded based on D2D and business |
| CN108629630A (en) * | 2018-05-08 | 2018-10-09 | 广州太平洋电脑信息咨询有限公司 | A kind of feature based intersects the advertisement recommendation method of joint deep neural network |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112402978A (en) * | 2020-11-13 | 2021-02-26 | 上海幻电信息科技有限公司 | Map generation method and device |
| CN114666655A (en) * | 2022-03-15 | 2022-06-24 | 北京达佳互联信息技术有限公司 | Video processing method and device, electronic equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109413149B (en) | 2021-05-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109902849B (en) | User behavior prediction method and device, and behavior prediction model training method and device | |
| CN101013427B (en) | Method and system for managing data | |
| US8504556B1 (en) | System and method for diminishing workload imbalance across multiple database systems | |
| CN103329151B (en) | Recommendation based on topic cluster | |
| CN109982104B (en) | Motion-aware video prefetching and cache replacement decision method in motion edge calculation | |
| CN110168529A (en) | Date storage method, device and storage medium | |
| CN110765353A (en) | Processing method and device of project recommendation model, computer equipment and storage medium | |
| CN109358821A (en) | A kind of cold and hot data store optimization method of cloud computing of cost driving | |
| CN113869801A (en) | Maturity state evaluation method and device for enterprise digital middleboxes | |
| RU2492522C2 (en) | System and method for efficient network simulation | |
| CN103140850B (en) | Data sharing in mobile device | |
| CN109413149A (en) | Information distribution control method, system, server and computer readable storage medium | |
| CN111935025B (en) | Control method, device, equipment and medium for TCP transmission performance | |
| CN108173958A (en) | Data-optimized storage method based on ant group algorithm under a kind of cloudy environment | |
| Zhao et al. | Integrating deep reinforcement learning with pointer networks for service request scheduling in edge computing | |
| US11184453B2 (en) | Systems and methods for managing content in a network | |
| CN110990706B (en) | Corpus recommendation method and device | |
| CN116681463A (en) | Sales prediction method, sales prediction system, sales prediction device and storage medium | |
| CN104123307A (en) | Data loading method and system | |
| Sharma et al. | Federated learning based caching in fog computing for future smart cities | |
| Hoiles et al. | Risk-averse caching policies for YouTube content in femtocell networks using density forecasting | |
| CN111435381A (en) | Request distribution method and device | |
| CN115907056A (en) | Prediction model training method, information prediction method and corresponding devices | |
| CN104636209A (en) | Resource scheduling system and method based on big data and cloud storage system directional performance optimization | |
| CN113538031B (en) | Training of multimedia resource allocation model and multimedia resource allocation 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 |