CN105760449A - Multi-source heterogeneous data cloud pushing method - Google Patents

Abstract

The invention discloses a multi-source heterogeneous data cloud pushing method. Based on the characteristics of multi-source heterogeneous data and the safety and privacy of mobile internet, the eigenvalue and eigenvector of multi-source heterogeneous data in a distributed environment are calculated to be used for quick separation of isomorphic data and heterogeneous data in a data source, and automatic separation and efficient pushing of isomorphic data and heterogeneous data are achieved with the cloud pushing technique. By designing a multi-dimensional decision-making cloud pushing model, synchronous data volume during data pushing and updating is reduced, time is shortened, cross-platform pushing of multi-source heterogeneous data is achieved, and work efficiency is improved remarkably especially under the condition that data volume is large, bandwidth is small and network is not reliable.

Description

A kind of cloud method for pushing towards multi-source heterogeneous data

Technical field

The present invention relates to a kind of cloud method for pushing towards multi-source heterogeneous data

Background technology

Along with the fast development of mobile communication technology, mobile Internet has become the main flow of internet development, becomes people and obtains the main channel of information.By mobile Internet, a large amount of multi-source heterogeneous Data Integrations create global information sharing environment together.In order to meet the demand of user, various Mobile solution are explosive increase, it is necessary to the data of propelling movement are also to multi-sourcing development.Catching of information on mobile terminal device is proposed high requirement by these multi-source heterogeneous data, is mainly reflected in mobility and time-bounded two aspects.Mobility requires message transmission when low-power consumption, low rate；Time-bounded require that information is sent on mobile terminal device at the appointed time.

How to push multi-source heterogeneous data fast and effectively and become the new problem that society faces.These multi-source heterogeneous data be difficult to share push also without unified method, traditional data push method has not adapted to the present user real-time requirement to information yet.Propelling movement mode currently mainly is actively the information of change to be sent to user by server, it is not necessary to user participates in, and decreases mutual number of times and burden, shortens the response time, improves efficiency, but these methods are not all suitable for the propelling movement of multi-source heterogeneous data.Although IOS and Android platform have the supplying system of oneself, but due at network, the restriction of operating system and application aspect, certain limitation is had on using, Google's cloud messenger service also cannot use at home, APNS on iPhone is also only applicable to iOS, it is impossible to cross-platform propelling movement, so also cannot be carried out the propelling movement of real-time high-efficiency both at home and abroad at present for cross-platform multi-source heterogeneous data.

Chinese scholars and research institution have been studied from different visual angles to multi-source heterogeneous data and propelling movement mode.First, a kind of algorithm obtaining high in the clouds sharing data stage by stage is devised from representative work mainly YangWang, the BharadwajVeeravall of data transfer direction, it is possible to the effective transmission cost controlling data.Domestic Xu Fulong, Liu Ming et al. further provides a kind of dynamic data transmission strategy based on relative distance perception, sensor node is adopted to come the size of computing node transmission probability the foundation of selection down hop when transmitting in this, as message to the relative distance of convergent point.

YangFang-Chun is had from the research in decision-making direction, SuSen proposes the Optimization Selection Algorithm of a kind of semantic Web service combination based on Fuzzy Multiple Attribute Decision Making Theory, this algorithm can evaluate the information described with real number, interval number and language type data, thus carrying out integrated decision-making.Jiang Qianyue, the Zhang Yaying of Tongji University further provides a kind of server push delivery method based on Fuzzy Synthetic Decision, and traditional long polling technique and poll are combined by the method, obtains a kind of combined type polling technique.

Realize in systematic research utilizing push technology, the Liu Xin of the Institute of Software, Chinese Academy of Science, Chen Wei proposes a kind of web tree assembly based on AJAX and ServerPush, provided the user be similar in windows explorer to directory tree operation basic function and Consumer's Experience.

But above method all realizes data mapping is pushed simply by amendment propelling movement mode, too much do not consider the propelling movement problem of multi-source heterogeneous data.

Summary of the invention

In order to overcome existing method for pushing cannot solve the propelling movement problem of multi-source heterogeneous data, the present invention proposes a kind of cloud method for pushing towards multi-source heterogeneous data, the method is for the feature of multi-source heterogeneous data, the features such as comprehensive mobile interchange software safety and privacy, calculate eigenvalue and the characteristic vector of multi-source heterogeneous data in distributed environment, for the isomorphism data in sharp separation data source and isomeric data, utilize cloud push technology to realize being automatically separated and efficiently pushing of isomorphism data and isomeric data.

Following technical scheme is provided in order to solve above-mentioned technical problem.

A kind of cloud method for pushing towards multi-source heterogeneous data, comprises the steps:

The first step: design cloud pushes platform, and process is as follows:

1.1 clouds push platform and push the message of three types: notice, transparent transmission message and Rich Media, notify as pushing the notification message being presented in notifications hurdle to mobile terminal；Transparent transmission message refers to, in the way of transparent transmission, self-defining content is sent to client, and developer can set rule in client in advance, carries out message customization；Rich Media refers to the information pushing the forms such as picture, video, audio frequency, network address；

1.2 clouds push platforms should support simultaneously to all users or according to labeling to specific user colony PUSH message, according to the message content that user subscribes to, by user grouping, to the label that grouping user is different, according to different labels, carry out the propelling movement of corresponding contents, unique user is carried out single label, finally pushes according to label substance；

1.3 clouds push platform and provide user profile and notification message statistical information, the feedback information statistical information retention ratio according to user, communication rate, flow consumption etc.；

1.4 clouds push platforms can cross-platform use (PC, iOS, Android), user can also according to oneself need add custom feature；

Second step: the collection of multi-source heterogeneous data, process is as follows:

Multi-source heterogeneous data in different system in environment are considered as cloud data by 2.1, and according to the different weights that system gives, just cloud data are categorized as different modules, such as notification module, Rich Media's module, transparent transmission message module；

2.2 clouds push the module information subscribed to according to user of platforms and are managed, and according to the module that different user is subscribed to, user are grouped, and having the user of equal modules is with group labelled.The information of the user of management registration simultaneously；

Cloud data message is managed by 2.3 cloud push server, information is managed according to the time, records the data of renewal every time, in order to use when next step pushes；

2.4 clouds push platform and the more fresh information in (2.3) are carried out pretreatment, and by identical for data form, what weights were identical is divided into different grouping, facilitates the separation of isomorphism data and isomeric data.The present invention proposes MDCP (MultipleDecisionCloudPush) model by the data of access to information database, is determined by weights and separates with attribute and carry out decision-making propelling movement, is whole " subscription-collection-decision-making-propelling movement " cycle；

3rd step: after information completes, it is necessary to cloud data are easily separated by MDCP model, is divided into multi-source heterogeneous data and multi-source isomorphism data, and process is as follows:

3.1 multi-source heterogeneous data component analyses

Due to being wanting in consideration to isomerism problem, cause the reduction of efficiency when pushing, even cannot work.So when in the face of isomeric data, it is necessary for selecting a suitable propelling movement mode.

The present invention needs the problem solved to be exactly utilize these to be distributed in the cloud data on different Cloud Server to obtain the main constituent of data, by each key data signal component value X_iThe deviant being this composition in the direction is subtracted each other, by the main component value X of each data with the μ provided_i(i=1,2...n) insert in matrix S.

S=[x₁,x₂,x₃...x_n](1)

Sue for peace after matrix S is multiplied by its diagonal matrix ST again and be averaged again, due to this Matrix Multiplication with obtain after its diagonal matrix for the numerical value determined, so this value can be used to represent overall isomery degree, represent with V.

$V=\frac{1}{N}\underset{i=1}{\overset{N}{\Σ}}\left\{\right|(x_i-\mathrm{\μ})\left|\right|{(x_i-\mathrm{\μ})}^T\left|\right\}---\left(2\right)$

Wherein μ is the weighted value drawn according to assessment, obtains its span according to different systems, and N is sample total, and owing to now V is the numerical value determined, the isomery degree available feature value V of this sample represents.

Needing to push if there is multi-source heterogeneous data, the now main constituent of each data can not be divided into X₁～X_N, now need X_iBecome X_ij, so matrix S becomes:

$S=\left[\begin{array}{c}{X}_{11},X_{12},...X_{1N}\\ X_{21},X_{22},...X_{2N}\\ ........................\\ X_{i1},X_{i2},\mathrm{....}{X}_{iN}\end{array}\right]---\left(3\right)$

Now, V is matrix, is considered as eigenmatrix, so V is multiplied by its diagonal matrix, can release the eigenvalue k of cloud data:

|V-kE|(4)

V gives characteristic vector, has been obtained size and the V common metrics isomery degree of different cloud data of corresponding eigenvalue k, eigenvalue k by abbreviation.

But the problem not accounting for data transfer load in the method, cloud data transmission different in distributed environment is subject to communication rate, the restriction of the factors such as computing capability, wish to reduce the data traffic between Cloud Server, avoid directly transmitting great amount of samples, data can be carried out permutation and combination in a matrix fashion, then carry out cloud propelling movement based on this data structure.

The propelling movement of 3.2 multi-source isomorphism data, process is as follows:

In cloud data, although there are substantial amounts of multi-source heterogeneous data, but there is also magnanimity multi-source isomorphism data, when the propelling movement content that user needs belongs to isomorphism data, use covariance matrix that data are pressed eigendecomposition, decompose isomorphism data and just can use more efficient propelling movement mode, it is not necessary to iterate at each point, saved data communication rates；

First, local variance and mean vector μ are estimated with the sample in first cloud data station₁, then obtain the sample estimation local variance in second cloud data station and mean vector μ₂, until obtaining the sample N of all cloud data station_μ。N_kμkFor the residue sample not updated, μ_KFor sample N_KMean vector.In the prescribed limit of local variance and mean vector, then it is assumed that be that these cloud data station provide isomorphism data, obtain average mean vector μ, again through N_μAnd N_kμkObtain mean parameter covariance:

$\stackrel{\‾}{\mathrm{\μ}}=\frac{1}{N+N_k}---\left(5\right)$

$\underset{i=1}{\overset{N+N_k}{\Σ}}{X}_i=\frac{1}{N+N_K}(N_{\mathrm{\μ}}+N_{K\mathrm{\μ}k})---\left(6\right)$

Therefore, when each cloud data sample is sufficient, the accuracy calculating isomorphism data is very big, and owing to being isomorphism data, secondary incremental update equation is:

$\Σ*=\frac{1}{N+N_k}\underset{i=1}{\overset{N+N_k}{\Σ}}(x_i-\mathrm{\μ}*){(x_i-\mathrm{\μ}*)}^T---\left(7\right)$

$\mathrm{\μ}*=\stackrel{\‾}{\mathrm{\μ}}-{\mathrm{\μ}}_k---\left(8\right)$

Wherein (K=1,2,3.....), due in above-mentioned algorithm, it may be noted that covariance matrix is symmetrical, therefore in propelling movement process, isomorphism data update every time, having only to half covariance matrix of transmission, MDCP model uses MQTT (MessageQueuingTelemetryTransport, the message queue remote measurement transmission of sing on web Socket, it is an instant communication protocol of IBM exploitation) push, if number of systems is M, number of parameters is d, and the time complexity of this algorithm is:

O (M, d)=(M-1) (d (d+1)/2+d+1)

The propelling movement of 3.3 multi-source heterogeneous data

Multi-source heterogeneous data are that isomorphism data main component is extended, and in order to facilitate mathematical calculation, introduce permutation matrix P, the sample that Cloud Server obtains is carried out displacement and maps, and result is designated as y, it may be assumed that

Y=(y_a,y_b)^t=P_mX(9)

Its objective is, by current sample focuses on before vector with identical part in isomorphism data, to use y_aRepresent, and different are partially disposed in after vector, use y_bRepresenting, equally, Mean Matrix and covariance matrix being replaced, result is designated as μ respectively, ∑:

μ=(μ_a,μ_b)^t=P_mμ

$\mathrm{\Σ}=\left(\begin{array}{c}\Σa\Σb\\ {\Σ}_b^t\Σc\end{array}\right)=Pm\Σ{P_m}^T---\left(10\right)$

μ and permutation matrix are all the amounts calculated the last time, and Σ a is the covariance matrix obtained in m sample, and Σ c is and the covariance matrix of different elements in isomorphism data, and Σ b is the matrix that between them, covariance is constituted.

This formula meets higher-dimension distribution, so cloud data sample just can adopt the distribution of this higher-dimension to calculate, its expression formula is as follows, and d is the dimension values in higher-dimension distribution:

$\begin{array}{c}N\left(y\right|\mathrm{\μ},\Σ)=\frac{1}{{\left(2^{\mathrm{\π}}\right)}^{d/2}{\Σ}^{1/2}}\\ \exp \left\{\frac{1}{2}(y-\mathrm{\μ})T{\Σ}^{-1}(y-\mathrm{\μ})\right\}\end{array}---\left(11\right)$

Y, μ, ∑ is brought into by (5) (9) (10) 3 formula and is simplified, and takes natural logrithm and can obtain

$\begin{array}{c}l(\mathrm{\μ},\Σ|y)=\\ \underset{i}{\Σ}\log P({\mathrm{\μ}}_a,{\Σ}_a|y_{aj},\mathrm{\μ}b,\Σc,\Σb)\end{array}---\left(12\right)$

Seek local derviation again, can obtain after abbreviation:

$\begin{array}{c}\Σa=\frac{1}{N_n}\underset{i=1}{\overset{Nm}{\Σ}}(y_a-{\mathrm{\μ}}_a){(y_a-{\mathrm{\μ}}_a)}^T-\\ {\Σ}_c^{-1}{\Σ}_b^T\frac{1}{N_M}\underset{i=1}{\overset{N_M}{\Σ}}(y_b-{\mathrm{\μ}}_b){(y_b-{\mathrm{\μ}}_b)}^T\\ {\Σ}_c^{-1}\Σb+\Σb{\Σ}_c^{-1}+\Σb{\Σ}_c^{-1}{\Σ}_b^T\end{array}---\left(13\right)$

Wherein the first row is the element entry identical with isomorphism data, calculates in formula meter (10), and the second row is exactly isomeric data, so information source pushes isomeric data formula after updating is:

$\Σa=\frac{1}{N}\underset{i=1}{\overset{N}{\Σ}}(y_b-{\mathrm{\μ}}_b){(y_b-{\mathrm{\μ}}_b)}^T---\left(14\right)$

Owing to, in isomeric data, the number of the element of every kind of cloud data is dissimilar, so requiring to look up each N, due to the existence of transposed matrix, it is only necessary to the data of transmission half.What therefore transmission was complicated is:

O (M, d)=(M-1) (d²+2d)

The isomeric data of this some of complex pushes, MDCP model will call on the Internet disclosed in third party's free cloud Push Service (such as Baidu's cloud propelling movement etc.) of increasing income push.The message push service of third-party platform is completely free, it is possible to zero cost uses, and has powerful server cluster, has high handling capacity, and the message that user subscribes to can send to user side at faster speed.

3.4MDCP model running flow process

By above-mentioned analysis, original supplying system is really difficult to the real-time high-efficiency to multi-source heterogeneous data and pushes, and after obtaining the task of needing to push, calls MDCP model.

When message enters and pushes list, MDCP model isolates multi-source isomorphism data and multi-source heterogeneous data by eigenvalue calculation, first cloud data are carried out the determination of weights, the size of weights is allocated according to the significance level of attribute by system, then is easily separated according to weights by attribute, if multi-source isomorphism data, third party is then used to push mode, if multi-source heterogeneous data then use MQTT to push, if cannot push, then reenter wait passage.

Cyclicity according to this cycle, it is used for MDCP model filtering out repetition data and delta data information, multi-source isomorphism data or multi-source heterogeneous data are distinguished further according to characteristic vector, quantity finally according to the system that user subscribes to, the bandwidth of network, the propelling movement mode that the comprehensive descision such as the quantity of the data of propelling movement and confidentiality uses.

Further, it is the logical architecture pushing platform towards multi-source heterogeneous data cloud that described cloud pushes framework, is divided into following five layers:

4.1 cloud data Layers:

This layer is the major part that cloud pushes in platform, and the multi-source heterogeneous data in each system define cloud data, and all data broadly fall into this layer, it is achieved that the logical Virtual of data, and the multi-source heterogeneous data gathered in second step are from this layer.

4.2 data management layer:

This layer is the core of framework, it is achieved that data and the collaborative work of propelling movement in cloud propelling movement, externally also provides for the access service with identical data.This layer is connected with cloud data Layer, after data acquisition completes, is completed the selection pushed by MDCP model, carries out the packet of data simultaneously according to application-interface layer.

4.3 application-interface layers

This layer can on-demand configure, according to the different application of user's subscription or module, it is provided that different pieces of information storage and the application service accessed, such as wage Push Service, school grade service, information service of speculating in shares, video monitoring service etc., these services are mutual by application-interface layer and cloud data Layer.

4.4 access layers:

Authorized user all can access cloud supplying system by the interface of standard, according to the different user types in client layer, gives different user rights, and the mode that all types of user accesses is also incomplete same.

4.5 client layers:

Client layer primary responsibility manages all kinds of registration users, also is responsible for the amendment of user profile simultaneously, updates, inquiry etc., also manages all kinds of registration terminal simultaneously, as: PC, IOS, Android etc..

The cloud method for pushing of present invention design, towards multi-source heterogeneous data, comes sharp separation isomorphism data and isomeric data by the eigenvalue and characteristic vector calculating data, to realize the efficient propelling movement of data.(MDCPMultipledecisioncloudpush) model is pushed by designing Multidimensional decision-making cloud, cloud is made to push the synchrodata amount not only reducing in data-pushing renewal process, shorten the time, and the problem solving the cross-platform propelling movement of multi-source heterogeneous data, huge particular for data volume, and it being operated in low bandwidth, in the insecure situation of network, work efficiency is obviously improved.

The present invention utilizes multi-source heterogeneous data cloud push technology, solves the deficiency in current supplying system.Compared with prior art have the following advantages:

1. it is applicable to the propelling movement occasion that requirement of real-time is high

2. according to the label of user property, packet test push function can be analyzed, carries out intelligently pushing

3. push process operation simple, precise and high efficiency.

4. cloud push server real-time perception data are abnormal, and when propelling data changes, server active push information, according to APP end, embodies the efficient of propelling movement, in real time.

Accompanying drawing explanation

Fig. 1 is the schematic diagram that cloud pushes platform feature design.

Fig. 2 is the schematic diagram of MDCP model running process.

Fig. 3 is the schematic diagram that cloud pushes Process Design.

Fig. 4 is the schematic diagram that cloud pushes logical architecture.

Fig. 5 is communication rate variation diagram.

Fig. 6 is cloud method for pushing flow diagram.

Fig. 7 is the schematic diagram that tradition pushes platform retention ratio.

Fig. 8 is the schematic diagram that cloud pushes platform retention ratio.

Detailed description of the invention

Below in conjunction with accompanying drawing, the present invention will be further described.

Embodiment 1

With reference to Fig. 1～Fig. 8, a kind of cloud method for pushing towards multi-source heterogeneous data, comprise the following steps:

The first step: design cloud pushes platform, and each functional module in platform is resolved, and cloud pushes platform feature and designs as shown in Figure 1

1.1 clouds push platform can push the message of three types: notice, transparent transmission message and Rich Media.Notice for pushing the notification message being presented in notifications hurdle to mobile terminal.Transparent transmission message refers to, in the way of transparent transmission, self-defining content is sent to client.Developer can set rule in client in advance, carries out message customization.Rich Media refers to the information pushing the forms such as picture, video, audio frequency, network address.

1.2 clouds push platforms should support simultaneously to all users or according to labeling to specific user colony PUSH message, according to the message content that user subscribes to, by user grouping, to the label that grouping user is different, according to different labels, carry out the propelling movement of corresponding contents, unique user is carried out single label, finally pushes according to label substance.

1.3 clouds push platform and provide user profile and notification message statistical information, the feedback information statistical information retention ratio according to user, communication rate, flow consumption etc.

1.4 clouds push platforms can cross-platform use (PC, iOS, Android), user can also according to oneself need add custom feature.

Second step: the gatherer process of multi-source heterogeneous data, as shown in Figure 3

Multi-source heterogeneous data in other each systems are considered as cloud data by 2.1, according to the information that user subscribes to, cloud data are categorized as different modules.

The module information that user is subscribed to by 2.2 clouds propelling movement platforms is managed, including subscribed content, log-on message, amendment information etc..

Cloud data message is managed by 2.3 cloud push server, including the renewal of cloud data, amendment etc..

2.4 clouds push platform and the more fresh information in (2.3) are carried out pretreatment, including the form of data, size etc..

3rd step: after information completes, it is necessary to cloud data are easily separated by MDCP model, is divided into multi-source heterogeneous data and multi-source isomorphism data, its flow process as shown in Figure 3:

3.1 multi-source heterogeneous data component analyses

The present invention needs the problem solved to be exactly utilize these to be distributed in the cloud data on different Cloud Server to obtain the main constituent of data, by each key data signal component value X_iThe deviant being this composition in the direction is subtracted each other, by the main component value X of each data with the μ provided_i(i=1,2...n) insert in matrix S.

S=[x₁,x₂,x₃...x_n](1)

Sue for peace after matrix S is multiplied by its diagonal matrix ST again and be averaged again, due to this Matrix Multiplication with obtain after its diagonal matrix for the numerical value determined, so this value can be used to represent overall isomery degree, represent with V.

$V=\frac{1}{N}\underset{i=1}{\overset{N}{\Σ}}\left\{\right|(x_i-\mathrm{\μ})\left|\right|{(x_i-\mathrm{\μ})}^T\left|\right\}---\left(2\right)$

Wherein μ is the weighted value drawn according to assessment, obtains its span according to different systems, and N is sample total, and owing to now V is the numerical value determined, the isomery degree available feature value V of this sample represents.

Needing to push if there is multi-source heterogeneous data, the now main constituent of each data can not be divided into X₁～X_N, now need X_iBecome X_IJ, so matrix S becomes:

$S=\left[\begin{array}{c}{X}_{11},X_{12},...X_{1N}\\ X_{21},X_{22},...X_{2N}\\ ........................\\ X_{i1},X_{i2},\mathrm{....}{X}_{iN}\end{array}\right]---\left(3\right)$

Now, V is matrix, is considered as eigenmatrix, so V is multiplied by its diagonal matrix, can release the eigenvalue k of cloud data:

|V-kE|(4)

V gives characteristic vector, has been obtained size and the V common metrics isomery degree of different cloud data of corresponding eigenvalue k, eigenvalue k by abbreviation.

But the problem not accounting for data transfer load in the method, cloud data transmission different in distributed environment is subject to communication rate, the restriction of the factors such as computing capability, wish to reduce the data traffic between Cloud Server, avoid directly transmitting great amount of samples, data can be carried out permutation and combination in a matrix fashion, then carry out cloud propelling movement based on this data structure.

The propelling movement of 3.2 multi-source isomorphism data

In cloud data, although there are substantial amounts of multi-source heterogeneous data, but there is also magnanimity multi-source isomorphism data, when the propelling movement content that user needs belongs to isomorphism data, use covariance matrix that data are pressed eigendecomposition, decompose isomorphism data and just can use more efficient propelling movement mode, it is not necessary to iterate at each point, saved data communication rates.

First, local variance and mean vector μ are estimated with the sample in first cloud data station₁, then obtain the sample estimation local variance in second cloud data station and mean vector μ₂, until obtaining the sample N of all cloud data station_μ。N_kμkFor the residue sample not updated, μ_KFor sample N_KMean vector.In the prescribed limit of local variance and mean vector, then it is assumed that be that these cloud data station provide isomorphism data, obtain average mean vector μ, again through N_μAnd N_kμkObtain mean parameter covariance:

$\stackrel{\‾}{\mathrm{\μ}}=\frac{1}{N+N_k}---\left(5\right)$

$\underset{i=1}{\overset{N+N_k}{\Σ}}{X}_i=\frac{1}{N+N_K}(N_{\mathrm{\μ}}+N_{K\mathrm{\μ}k})---\left(6\right)$

Therefore, when each cloud data sample is sufficient, the accuracy calculating isomorphism data is very big, and owing to being isomorphism data, secondary incremental update equation is:

$\Σ*=\frac{1}{N+N_k}\underset{i=1}{\overset{N+N_k}{\Σ}}(x_i-\mathrm{\μ}*){(x_i-\mathrm{\μ}*)}^T---\left(7\right)$

$\mathrm{\μ}*=\stackrel{\‾}{\mathrm{\μ}}-{\mathrm{\μ}}_k---\left(8\right)$

Wherein (K=1,2,3.....).Due in above-mentioned algorithm, it may be noted that covariance matrix is symmetrical, therefore in propelling movement process, isomorphism data update every time, having only to half covariance matrix of transmission, MDCP model uses the MQTT of sing on web Socket to push, if number of systems is M, number of parameters is d, and the time complexity of this algorithm is:

O (M, d)=(M-1) (d (d+1)/2+d+1)

The propelling movement of 3.3 multi-source heterogeneous data

Multi-source heterogeneous data are that isomorphism data main component is extended, and in order to facilitate mathematical calculation, introduce permutation matrix P, the sample that Cloud Server obtains is carried out displacement and maps, and result is designated as y, it may be assumed that

Y=(y_a,y_b)^t=P_mX(9)

Its objective is, by current sample focuses on before vector with identical part in isomorphism data, to use y_aRepresent, and different are partially disposed in after vector, use y_bRepresenting, equally, Mean Matrix and covariance matrix being replaced, result is designated as μ respectively, ∑:

μ=(μ_a,μ_b)^t=P_mμ

$\mathrm{\Σ}=\left(\begin{array}{c}\Σa\Σb\\ {\Σ}_b^t\Σc\end{array}\right)=Pm\Σ{P_m}^T---\left(10\right)$

μ and permutation matrix are all the amounts calculated the last time, and Σ a is the covariance matrix obtained in m sample, and Σ c is and the covariance matrix of different elements in isomorphism data, and Σ b is the matrix that between them, covariance is constituted.

This formula meets higher-dimension distribution, so cloud data sample just can adopt the distribution of this higher-dimension to calculate, its expression formula is as follows:

$\begin{array}{c}N\left(y\right|\mathrm{\μ},\Σ)=\frac{1}{{\left(2^{\mathrm{\π}}\right)}^{d/2}{\Σ}^{1/2}}\\ \exp \left\{\frac{1}{2}(y-\mathrm{\μ})T{\Σ}^{-1}(y-\mathrm{\μ})\right\}\end{array}---\left(11\right)$

Y, μ, ∑ is brought into by (5) (9) (10) 3 formula and is simplified, and takes natural logrithm and can obtain

$\begin{array}{c}l(\mathrm{\μ},\Σ|y)=\\ \underset{i}{\Σ}\log P({\mathrm{\μ}}_a,{\Σ}_a|y_{aj},\mathrm{\μ}b,\Σc,\Σb)\end{array}---\left(12\right)$

Seek local derviation again, can obtain after abbreviation:

$\begin{array}{c}\Σa=\frac{1}{N_n}\underset{i=1}{\overset{Nm}{\Σ}}(y_a-{\mathrm{\μ}}_a){(y_a-{\mathrm{\μ}}_a)}^T-\\ {\Σ}_c^{-1}{\Σ}_b^T\frac{1}{N_M}\underset{i=1}{\overset{N_M}{\Σ}}(y_b-{\mathrm{\μ}}_b){(y_b-{\mathrm{\μ}}_b)}^T\\ {\Σ}_c^{-1}\Σb+\Σb{\Σ}_c^{-1}+\Σb{\Σ}_c^{-1}{\Σ}_b^T\end{array}---\left(13\right)$

Wherein the first row is the element entry identical with isomorphism data, calculates in formula meter (10), and the second row is exactly isomeric data, so information source pushes isomeric data formula after updating is:

$\Σa=\frac{1}{N}\underset{i=1}{\overset{N}{\Σ}}(y_b-{\mathrm{\μ}}_b){(y_b-{\mathrm{\μ}}_b)}^T---\left(14\right)$

Owing to, in isomeric data, the number of the element of every kind of cloud data is dissimilar, so requiring to look up each N, due to the existence of transposed matrix, it is only necessary to the data of transmission half.What therefore transmission was complicated is:

O (M, d)=(M-1) (d²+2d)

The isomeric data of this some of complex pushes, MDCP model will call on the Internet disclosed in third party's free cloud Push Service (such as Baidu's cloud propelling movement etc.) of increasing income push.The message push service of third-party platform is completely free, it is possible to zero cost uses, and has powerful server cluster, has high handling capacity, and the message that user subscribes to can send to user side at faster speed.

3.4MDCP model running flow process

By above-mentioned analysis, original supplying system is really difficult to the real-time high-efficiency to multi-source heterogeneous data and pushes, and after obtaining the task of needing to push, calls MDCP model, and its running is as shown in Figure 2.When message enters and pushes list, MDCP model isolates multi-source isomorphism data and multi-source heterogeneous data by eigenvalue calculation, if multi-source isomorphism data, then use MQTT protocol propelling (Fig. 2 is 1.), if multi-source heterogeneous data then use third party to push mode (Fig. 2 is 2.), if cannot push, then reenter wait passage (Fig. 2 is 3.).

4th step: it is the logical architecture figure pushing platform towards multi-source heterogeneous data cloud that cloud pushes framework, and as shown in Figure 4, cloud pushes platform logic framework and generally can be divided into following five layers:

4.1 cloud data Layers:

This layer is the major part that cloud pushes in platform, and the multi-source heterogeneous data in each system define cloud data, and all data broadly fall into this layer, it is achieved that the logical Virtual of data, and the multi-source heterogeneous data gathered in second step are from this layer.

4.2 data management layer:

This layer is the core of framework, it is achieved that data and the collaborative work of propelling movement in cloud propelling movement, externally also provides for the access service with identical data.This layer is connected with cloud data Layer, after data acquisition completes, is completed the selection pushed by MDCP model, carries out the packet of data simultaneously according to application-interface layer.

4.3 application-interface layers

This layer can on-demand configure, according to the different application of user's subscription or module, it is provided that different pieces of information storage and the application service accessed, such as wage Push Service, school grade service, information service of speculating in shares, video monitoring service etc., these services are mutual by application-interface layer and cloud data Layer.

4.4 access layers:

Authorized user all can access cloud supplying system by the interface of standard, according to the different user types in client layer, gives different user rights, and the mode that all types of user accesses is also incomplete same.

4.5 client layers:

Client layer primary responsibility manages all kinds of registration users, also is responsible for the amendment of user profile simultaneously, updates, inquiry etc., also manages all kinds of registration terminal simultaneously, as: PC, IOS, Android etc..

Embodiment 2

The invention provides a kind of cloud method for pushing towards multi-source heterogeneous data, be applied to Android and iOS two large platform, comprise the following steps:

The first step: cloud method for pushing is used on Android and iOS platform.

1.1 present invention use cloud to push platform and are tested in Android and iOS, cloud Data Source is in result system, pay system and lithographic viewing system, after collecting data by the main component value weight of data it is: user right, user's login time, the pattern of user's publish/subscribe (one-to-many/one to one), user logs in quantity, transmitted data amount.Weight mu is determined through considering, it is assumed that in native system, signal component value weight is (0.3,0.2,0.2,0.1,0.1,0.1).Weight is more high, and the change of this signal component value is more big on the last MDCP impact pushing result.Owing to Data Source is in three systems, and the data type different (there is rich media information) of lithographic viewing system, so the data collected belong to multi-source heterogeneous data, as shown in table 1.

Table 1 cloud Data Data signal component value

Data in table one are brought in specific embodiment 1 and can be obtained in formula (3) by 1.2:

$S=\left[\begin{array}{cccccc}0.085& 0.122& 0.231& 0.081& 0.098& 0.95\\ 0.127& 0.128& 0.246& 0.200& 0.310& 0.810\\ 0.891& 0.129& 0.871& 0.764& 0.990& 0.769\\ 0.361& 0.765& 0.560& 0.809& 0.591& 0.760\\ 0.819& 0.581& 0.129& 0.794& 0.899& 0.870\\ 0.270& 0.800& 0.531& 0.750& 0.999& 0.969\end{array}\right]$

Matrix S and weighted value μ is substituted in specific embodiment 1 formula (2) can obtain:

$V_1=\left[\begin{array}{c}\left|0.085-0.3\right|\left|0.122-0.2\right|\left|0.231-0.2\right|\left|0.081-0.1\right|\left|0.098-0.1\right|\left|0.95-0.1\right|\\ \left|0.127-0.3\right|\left|0.128-0.2\right|\left|0.246-0.2\right|\left|0.200-0.1\right|\left|0.310-0.1\right|\left|0.81-0.1\right|\\ \left|0.891-0.3\right|\left|0.129-0.2\right|\left|0.871-0.2\right|\left|0.764-0.1\right|\left|0.990-0.1\right|\left|0.769-0.1\right|\\ \left|0.361-0.3\right|\left|0.765-0.2\right|\left|0.560-0.2\right|\left|0.809-0.1\right|\left|0.591-0.1\right|\left|0.760-0.1\right|\\ \left|0.819-0.3\right|\left|0.581-0.2\right|\left|0.129-0.2\right|\left|0.794-0.1\right|\left|0.899-0.1\right|\left|0.870-0.1\right|\\ \left|0.270-0.3\right|\left|0.800-0.2\right|\left|0.531-0.2\right|\left|0.750-0.1\right|\left|0.999-0.1\right|\left|0.969-0.1\right|\end{array}\right]$

$V=\frac{1}{5}\underset{i=1}{\overset{5}{\Σ}}{V}_1\·{V_1}^T=\left[\begin{array}{cccccc}0.012& 0.167& 0.231& 0.431& 0.981& 0.993\\ 0.125& 0.145& 0.246& 0.265& 0.230& 0.231\\ 0.801& 0.120& 0.871& 0.564& 0.923& 0.928\\ 0.323& 0.765& 0.523& 0.857& 0.391& 0.125\\ 0.832& 0.581& 0.127& 0.756& 0.867& 0.849\\ 0.246& 0.800& 0.531& 0.750& 0.012& 0.980\end{array}\right]$

After obtaining eigenmatrix, eigenvalue can be obtained by formula (4) in specific embodiment 1

$\left[\begin{array}{cccccc}0.012& 0.167& 0.231& 0.431& 0.981& 0.993\\ 0.125& 0.145& 0.246& 0.265& 0.230& 0.231\\ 0.801& 0.120& 0.871& 0.564& 0.923& 0.928\\ 0.323& 0.765& 0.523& 0.857& 0.391& 0.125\\ 0.832& 0.581& 0.127& 0.756& 0.867& 0.849\\ 0.246& 0.800& 0.531& 0.750& 0.012& 0.980\end{array}\right]-k\left[\begin{array}{c}100000\\ 010000\\ 001000\\ 000100\\ 000010\\ 000001\end{array}\right]$

Can obtain after abbreviation:

K1=1.2k2=1.6k3=4.9

K3=4.9 substantially belongs to isomeric data, k1, and k2 can be considered as multi-source isomorphism data.

$\begin{array}{cc}{y}_a=\left[\begin{array}{cccc}0.085& 0.122& 0.231& 0.081\\ 0.127& 0.128& 0.246& 0.200\\ 0.891& 0.129& 0.871& 0.764\\ 0.361& 0.765& 0.560& 0.809\\ 0.819& 0.581& 0.129& 0.794\\ 0.270& 0.800& 0.531& 0.750\end{array}\right]& y_b=\left[\begin{array}{cc}0.098& 0.95\\ 0.310& 0.810\\ 0.990& 0.769\\ 0.591& 0.760\\ 0.899& 0.870\\ 0.999& 0.969\end{array}\right]\end{array}$

$y={(y_a,y_b)}^t=\left[\begin{array}{cccccc}0.085& 0.127& 0.981& 0.361& 0.819& 0.270\\ 0.122& 0.128& 0.129& 0.765& 0.581& 0.800\\ 0.231& 0.246& 0.871& 0.560& 0.129& 0.531\\ 0.081& 0.200& 0.764& 0.809& 0.794& 0.750\\ 0.098& 0.310& 0.990& 0.591& 0.899& 0.999\\ 0.950& 0.810& 0.769& 0.760& 0.870& 0.969\end{array}\right]$

1.3

The data component value average of each time period of table 2

As shown in Table 2, in order to try to achieve Mean Matrix μ, measure 6 class means by 6 different time sections, constitute matrix μ:

$\mathrm{\μ}=\left[\begin{array}{cccccc}0.05& 0.08& 0.085& 0.09& 0.095& 0.10\\ 0.120& 0.139& 0.184& 0.618& 0.200& 0.08\\ 0.641& 0.149& 0.140& 0.210& 0.300& 0.20\\ 0.2003& 0.09& 0.051& 0.230& 0.201& 0.90\\ 0.230& 0.12& 0.232& 0.560& 0.640& 0.80\\ 0.917& 0.60& 0.090& 0.150& 0.150& 0.78\end{array}\right]$

MDCP finally determines comprehensive propelling movement mode, owing to weight is k=(0.3,0.2,0.2,0.1,0.1), thus l (μ, Σ y)=Σ | y k

μ, Σ y is substituted into specific embodiment 1 formula (12), draws higher-dimension distribution N=(L, μ, Σ), calculate N=(0.35,0.25,0.15) according to vector.L is substituted in specific embodiment 1 formula (13), it can be deduced that:

$\Σa=\left[\begin{array}{c}0.038,0.050,0.500,0.020,0,0\\ 0.050,0.040,0.010,0,0.750,0\\ 0.500,0.010,0.028,0,0,0.980\end{array}\right]$

Owing to each humping section is the data updated, so removing 3 row redundant datas after in matrix, obtain Σ d for more new data:

$\Σd=\left[\begin{array}{c}0.038,0.050,0.500\\ 0.050,0.040,0.010\\ 0.500,0.010,0.028\end{array}\right]$

It can be seen that Σ d is symmetrical matrix from matrix, so updating is data volume is original tradition propelling movement mode 1/2nd of transmission every time, substantially increase pushing efficiency.

Second step: cloud pushes platform and provides user profile and notification message statistical information, the feedback information statistical information retention ratio according to user, communication rate, flow consumption etc.

2.1 communication rate analyses, as it is shown in figure 5, abscissa represents number of tasks, vertical coordinate represents communication rate

For weighing the ability of the MDCP algorithm that the present invention proposes, from average transmission rate, communication rate, flow during standing, push retention ratio and the model validation in multi-source heterogeneous data cloud method for pushing is estimated.

Communication rate refers to that in the unit interval, user and cloud push the number of communications of platform, whether test user is ready to use this platform to carry out data-pushing, and test simultaneously and can or can not produce other problems when number of communications is high, it is owing to all can send request to server in the whole process pushing task queue message always that traditional propelling movement mode communication rate is basically unchanged, and the cloud that the present invention proposes pushes model and is in information collecting step, operation along with system, task increasing number, advantage just shows gradually, more many in number of tasks, the traffic of cost tails off on the contrary, as shown in table 3.

Table 3 average transmission rate tables of data

2.2 cloud method for pushing flows, as shown in Figure 6, abscissa represents flow value, and vertical coordinate represents time of repose.

During standing, flow refers to the added flow produced when mobile phone stands due to propelling movement, cloud pushes whether platform can produce a large amount of flow because of the improvement of communication rate, using cloud to push platform respectively and traditional platform that pushes compares, result shows that cloud pushes platform and consumes less than tradition propelling movement platform in the mobile longer situation down-off of equipment time of repose.Average transmission rate refers to that, by repeatedly comparing calculation, with the transmission quantity that common propelling movement mode compares, result is as shown in table 4:

Current capacity contrast's table when table 3 stands

2.3 cloud method for pushing retention ratio, as shown in Figure 7 and Figure 8, abscissa represents that number of weeks, vertical coordinate represent the percentage ratio of retention ratio.

Retention ratio refers within a certain period of time, such as 1-6 week, user also leaves the ratio of this PUSH message, also can to a certain degree reflecting the impact on user of this model, tradition pushes platform and pushes platform retention ratio comparison diagram with cloud, result shows that the retention ratio of almost cloud per week propelling movement platform both is greater than tradition and pushes more than one times of platform retention ratio, and after 6 week, tradition pushes the retention ratio of platform less than 20%, and cloud propelling movement platform nearly reaches 50%.

Claims (2)

1. the cloud method for pushing towards multi-source heterogeneous data, it is characterised in that: described cloud method for pushing comprises the steps:

The first step: design cloud pushes platform, and process is as follows:

1.1 clouds push platform and push the message of three types: notice, transparent transmission message and Rich Media；

1.2 clouds push platforms should support simultaneously to all users or according to labeling to specific user colony PUSH message, according to the message content that user subscribes to, by user grouping, to the label that grouping user is different, according to different labels, carry out the propelling movement of corresponding contents, unique user is carried out single label, finally pushes according to label substance；

1.3 clouds push platform and provide user profile and notification message statistical information, the feedback information statistical information retention ratio according to user, communication rate and flow consumption；

1.4 clouds push the cross-platform uses of platform, user according to oneself need add custom feature；

Second step: the collection of multi-source heterogeneous data, process is as follows:

Multi-source heterogeneous data in different system in environment are considered as cloud data by 2.1, and according to the different weights that system gives, just cloud data are categorized as different modules, such as notification module, Rich Media's module and transparent transmission message module；

2.2 clouds push the module information subscribed to according to user of platforms and are managed, and according to the module that different user is subscribed to, user are grouped, and have the user of equal modules for group labelled, and the information of user is registered in management simultaneously；

Cloud data message is managed by 2.3 cloud push server, information is managed according to the time, records the data of renewal every time, in order to use when next step pushes；

2.4 clouds push platform and the more fresh information in (2.3) are carried out pretreatment, by identical for data form, what weights were identical is divided into different grouping, facilitate the separation of isomorphism data and isomeric data, MDCP model is by the data of access to information database, it is determined by weights to separate with attribute and carry out decision-making propelling movement, is whole " subscription-collection-decision-making-propelling movement " cycle；

3rd step: after information completes, it is necessary to cloud data are easily separated by MDCP model, is divided into multi-source heterogeneous data and multi-source isomorphism data, and process is as follows:

3.1 multi-source heterogeneous data component analyses

Utilize these to be distributed in the isomeric data on different Cloud Server to obtain the main constituent of data, subtract each other, with the μ provided, the deviant being this composition in the direction by each key data signal component value Xi, by the main component value X of each data_iInsert in matrix S, i=1,2...n；

S=[x₁,x₂,x₃...x_n](1)

Sue for peace after matrix S is multiplied by its diagonal matrix ST again and be averaged again, due to this Matrix Multiplication with obtain after its diagonal matrix for the numerical value determined, so this value is used for representing overall isomery degree, represent with V:

$V=\frac{1}{N}\underset{i=1}{\overset{N}{\Σ}}\left\{\right|(x_i-\mathrm{\μ})\left|\right|{(x_i-\mathrm{\μ})}^T\left|\right\}---\left(2\right)$

Wherein μ is the weighted value drawn according to assessment, obtains its span according to different systems, and N is sample total, and owing to now V is the numerical value determined, the isomery degree available feature value V of this sample represents；

Needing to push if there is multi-source heterogeneous data, the now main constituent of each data can not be divided into X₁～X_N, now need X_iBecome X_ij, so matrix S becomes:

$S=\left[\begin{array}{c}{X}_{11},X_{12},\mathrm{...}{X}_{1N}\\ X_{21},X_{22},\mathrm{...}{X}_{2N}\\ \mathrm{........................}\\ X_{i1},X_{i2},\mathrm{....}{X}_{iN}\end{array}\right]---\left(3\right)$

Now, V is matrix, is considered as eigenmatrix, so V is multiplied by its diagonal matrix, can release the eigenvalue k of cloud data:

|V-kE|(4)

V gives characteristic vector, has been obtained size and the V common metrics isomery degree of different cloud data of corresponding eigenvalue k, eigenvalue k by abbreviation；

The propelling movement of 3.2 multi-source isomorphism data, process is as follows:

First, local variance and mean vector μ are estimated with the sample in first cloud data station₁, then obtain the sample estimation local variance in second cloud data station and mean vector μ₂, until obtaining the sample N of all cloud data station_K, in the prescribed limit of local variance and mean vector, then it is assumed that be that these cloud data station provide isomorphism data, obtain average mean vector μ, again throughObtain mean parameter covariance:

$\stackrel{\‾}{\mathrm{\μ}}=\frac{1}{N+N_k}---\left(5\right)$

$\underset{i=1}{\overset{N+N_k}{\Σ}}{X}_i=\frac{1}{N+N_K}(N_{\mathrm{\μ}}+N_{{\mathrm{K\μ}}_k})---\left(6\right)$

Owing to being isomorphism data, secondary incremental update equation is:

$\mathrm{\Σ}*=\frac{1}{N+N_k}\underset{i=1}{\overset{N+N_k}{\Σ}}(x_i-\mathrm{\μ}*){(x_i-\mathrm{\μ}*)}^T---\left(7\right)$

$\mathrm{\μ}*=\stackrel{\‾}{\mathrm{\μ}}-{\mathrm{\μ}}_k---\left(8\right)$

Wherein, K=1,2,3....., in propelling movement process, isomorphism data update every time, it is only necessary to transmit half covariance matrix, and MDCP model uses the MQTT of sing on web Socket to push, if number of systems is M, number of parameters is d, and time complexity is:

O (M, d)=(M-1) (d (d+1)/2+d+1)；

The propelling movement of 3.3 multi-source heterogeneous data

Multi-source heterogeneous data are that isomorphism data main component is extended, and introduce permutation matrix P, the sample that Cloud Server obtains is carried out displacement and maps, and result is designated as y, it may be assumed that

Y=(y_a,y_b)^t=P_mX(9)

By current sample focuses on before vector with identical part in isomorphism data, use y_aRepresent, and different are partially disposed in after vector, use y_bRepresenting, equally, Mean Matrix and covariance matrix being replaced, result is designated as μ respectively, ∑:

$\begin{array}{c}\mathrm{\μ}={\left({\mathrm{\μ}}_a,{\mathrm{\μ}}_b\right)}^t=P_m\mathrm{\μ}\\ \mathrm{\Σ}=\left(\begin{array}{c}\mathrm{\Σ}a\mathrm{\Σ}b\\ {\mathrm{\Σ}}_b^t\mathrm{\Σ}c\end{array}\right)={\mathrm{Pm\ΣP}}_m^T\end{array}---\left(10\right)$

μ and permutation matrix are all the amounts calculated the last time, and Σ a is the covariance matrix obtained in m sample, and Σ c is and the covariance matrix of different elements in isomorphism data, and Σ b is the matrix that between them, covariance is constituted；

Cloud data sample adopts the distribution of this higher-dimension to calculate, and its expression formula is as follows:

$\begin{array}{c}N\left(y|\mathrm{\μ},\mathrm{\Σ}\right)=\frac{1}{{\left(2\mathrm{\π}\right)}^{d/2}{\mathrm{\Σ}}^{1/2}}\\ \exp \left\{\frac{1}{2}(y-\mathrm{\μ}){\mathrm{T\Σ}}^{-1}(y-\mathrm{\μ})\right\}\end{array}---\left(11\right)$

Y, μ, ∑ is brought into by (5) (9) (10) 3 formula and is simplified, and takes natural logrithm and obtain

$\begin{array}{c}l\left(\mathrm{\μ},\mathrm{\Σ}|y\right)=\\ \underset{i}{\mathrm{\Σ}}\log P\left({\mathrm{\μ}}_a,{\mathrm{\Σ}}_a|y_{aj},\mathrm{\μ}b,\mathrm{\Σ}c,\mathrm{\Σ}b\right)\end{array}---\left(12\right)$

Seek local derviation again, after abbreviation:

$\begin{array}{c}\mathrm{\Σ}a=\frac{1}{N_m}\underset{i=1}{\overset{Nm}{\mathrm{\Σ}}}\left(y_a-{\mathrm{\μ}}_a\right){\left(y_a-{\mathrm{\μ}}_a\right)}^T-\\ {\mathrm{\Σ}}_c^{-1}{\mathrm{\Σ}}_b^T\frac{1}{N_M}\underset{i=1}{\overset{N_M}{\mathrm{\Σ}}}\left(y_b-{\mathrm{\μ}}_b\right){\left(y_b-{\mathrm{\μ}}_b\right)}^T\\ {\mathrm{\Σ}}_c^{-1}\mathrm{\Σ}b+{\mathrm{\Σb\Σ}}_c^{-1}+{\mathrm{\Σb\Σ}}_c^{-1}{\mathrm{\Σ}}_b^T\end{array}---\left(13\right)$

Wherein the first row is the element entry identical with isomorphism data, calculates in formula meter (10), and the second row is exactly isomeric data, so information source pushes isomeric data formula after updating is:

$\mathrm{\Σ}a=\frac{1}{N}\underset{i=1}{\overset{N}{\Σ}}(y_b-{\mathrm{\μ}}_b){(y_b-{\mathrm{\μ}}_b)}^T---\left(14\right)$

3.4MDCP model running flow process,

After obtaining the task of needing to push, call MDCP model,

When message enters and pushes list, MDCP model isolates multi-source isomorphism data and multi-source heterogeneous data by eigenvalue calculation, first cloud data are carried out the determination of weights, the size of weights is allocated according to the significance level of attribute by system, then is easily separated according to weights by attribute, if multi-source isomorphism data, third party is then used to push mode, if multi-source heterogeneous data then use MQTT to push, if cannot push, then reenter wait passage；

Cyclicity according to this cycle, it is used for MDCP model filtering out repetition data and delta data information, multi-source isomorphism data or multi-source heterogeneous data are distinguished further according to characteristic vector, quantity finally according to the system that user subscribes to, the bandwidth of network, the propelling movement mode that the quantity of the data of propelling movement and confidentiality comprehensive descision use.

2. a kind of cloud method for pushing towards multi-source heterogeneous data as claimed in claim 1, it is characterised in that: it is the logical architecture pushing platform towards multi-source heterogeneous data cloud that described cloud pushes framework, is divided into following five layers:

4.1 cloud data Layers: the multi-source heterogeneous data in each system define cloud data, and all data broadly fall into this layer, it is achieved that the logical Virtual of data, the multi-source heterogeneous data gathered in second step are from this layer；

4.2 data management layer: realize the collaborative work of data and propelling movement during cloud pushes, externally also providing for the access service with identical data, this layer is connected with cloud data Layer, after data acquisition completes, completed the selection pushed by MDCP model, carry out the packet of data simultaneously according to application-interface layer；

4.3 application-interface layers: this layer can on-demand configure, the different application subscribed to according to user or module, it is provided that different pieces of information storage and the application service accessed, these services are mutual by application-interface layer and cloud data Layer；

4.4 access layers: authorized user all can access cloud supplying system by the interface of standard, according to the different user types in client layer, give different user rights, and the mode that all types of user accesses is also incomplete same；

4.5 client layers: be responsible for all kinds of registration user, also are responsible for the amendment of user profile simultaneously, update, inquiry etc., also manage all kinds of registration terminal simultaneously.