CN106209022B - A kind of filtering method and device - Google Patents

A kind of filtering method and device Download PDF

Info

Publication number
CN106209022B
CN106209022B CN201610554543.7A CN201610554543A CN106209022B CN 106209022 B CN106209022 B CN 106209022B CN 201610554543 A CN201610554543 A CN 201610554543A CN 106209022 B CN106209022 B CN 106209022B
Authority
CN
China
Prior art keywords
filtering
signal
signals
subnets
filtered
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.)
Active
Application number
CN201610554543.7A
Other languages
Chinese (zh)
Other versions
CN106209022A (en
Inventor
牛凯
乔雨倩
贺志强
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing University of Posts and Telecommunications
Original Assignee
Beijing University of Posts and Telecommunications
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Beijing University of Posts and Telecommunications filed Critical Beijing University of Posts and Telecommunications
Priority to CN201610554543.7A priority Critical patent/CN106209022B/en
Publication of CN106209022A publication Critical patent/CN106209022A/en
Application granted granted Critical
Publication of CN106209022B publication Critical patent/CN106209022B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • H—ELECTRICITY
    • H03—ELECTRONIC CIRCUITRY
    • H03H—IMPEDANCE NETWORKS, e.g. RESONANT CIRCUITS; RESONATORS
    • H03H17/00—Networks using digital techniques
    • H03H17/02—Frequency selective networks
    • H03H17/0201—Wave digital filters

Landscapes

  • Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Computer Hardware Design (AREA)
  • Mathematical Physics (AREA)
  • Data Exchanges In Wide-Area Networks (AREA)

Abstract

The embodiment of the present invention provides a kind of filtering method and device, it is related to filtering technique application field, wherein filtering method includes: the multiple subnets for obtaining heterogeneous network, and plurality of subnet includes the node of different predefined types, and each subnet is the network of same predefined type subnet node composition；Construct respectively connection in the subnet of multiple subnets between subnet node, multiple subnets two sub- gateway nodes between connection and multiple subnets hypergraph node between the corresponding tensor property data of connection, wherein hypergraph is what three or more subnets were formed；An ancestor node signal is configured for heterogeneous network；According to tensor property data, ancestor node signal is propagated on upper, hypergraph in subnet, between two subnets respectively, and correspond to the first filtering signal, multiple second filtering signals and the multiple third filtering signals after being propagated；According to the first filtering signal, multiple second filtering signals and multiple third filtering signals, the propagation filtering signal of heterogeneous network is determined.

Description

Filtering method and device
Technical Field
The present invention relates to the field of filtering technology application, and in particular, to a filtering method and apparatus.
Background
At present, high-dimensional data such as energy, transmission, neural networks and the like are researched on the basis of graph theory structures. The graphs in the graph theory structure are abstractions and generalizations of the various types of graphs, where the points of the graphs in the graph theory represent objects under study and the edges of the graphs in the graph theory represent the connections between the objects under study. A further way of describing the diagram is to set the signals on the diagram. For example, in a transmission network, signals may be used to describe the spread of pathogens, the migration of humans, or the movement of inventory.
In a social network, the relationships among users form a complex network, and user behavior data also contains a large amount of valuable information. Graph filtering algorithms may be used when jointly analyzing user behavior and relationships between users.
Matrix-based graph filtering in existing graph filtering algorithms. The graph filtering only comprises the structural characteristics of the complex network with one type of nodes, and the complex network with one type of nodes is filtered, so that the complex network with higher dimension (the complex network comprising a plurality of different types of nodes) cannot be analyzed, and the signals of the complex network can be filtered.
Disclosure of Invention
The embodiment of the invention aims to provide a filtering method and a filtering device, which comprehensively utilize different types of nodes of a heterogeneous network and the connection of the different types of nodes to filter original node signals of the heterogeneous network.
In order to achieve the above object, an embodiment of the present invention discloses a filtering method, wherein the filtering method includes the following steps:
acquiring a plurality of subnets of a heterogeneous network, wherein the subnets comprise nodes of different preset types, and each subnet is a network formed by subnet nodes of the same preset type;
constructing tensor characteristic data corresponding to the connection among the sub-network nodes in the sub-networks of the sub-networks, the connection among two sub-network nodes of the sub-networks and the connection among hypergraph nodes of the sub-networks respectively, wherein the hypergraph is formed by more than three sub-networks;
configuring an original node signal for the heterogeneous network;
according to the tensor characteristic data, the original node signals are respectively transmitted in the subnets, between the two subnets and on the hypergraph, and a first filtering signal, a plurality of second filtering signals and a plurality of third filtering signals after transmission are correspondingly obtained;
and determining a propagation filter signal of the heterogeneous network according to the first filter signal, the plurality of second filter signals and the plurality of third filter signals.
In order to achieve the above object, an embodiment of the present invention further discloses a filtering apparatus, including:
the device comprises an acquisition module, a processing module and a processing module, wherein the acquisition module is used for acquiring a plurality of subnets of a heterogeneous network, the subnets comprise nodes of different preset types, and each subnet is a network formed by subnet nodes of the same preset type;
the constructing module is used for respectively constructing tensor characteristic data corresponding to the connection among the subnet nodes in the subnets of the subnets, the connection among two nodes among the subnets of the subnets and the connection among nodes of the hypergraph of the subnets, wherein the hypergraph is formed by more than three subnets;
the configuration module is used for configuring an original node signal for the heterogeneous network;
a filtered signal obtaining module, configured to separately propagate the original node signal in the subnet, between the two subnets, and on the hypergraph according to the tensor feature data, and correspondingly obtain a propagated first filtered signal, a plurality of second filtered signals, and a plurality of third filtered signals;
and the processing module is used for determining the propagation filtering signals of the heterogeneous network according to the first filtering signals, the plurality of second filtering signals and the plurality of third filtering signals.
As can be seen from the above technical solutions, in the embodiments of the present invention, by obtaining subnets, tensor feature data corresponding to connections between subnet nodes in the subnets, connections between two subnet nodes, and connections between hypergraph nodes is constructed, and an original node signal is set for a heterogeneous network, and then the original node signal of the heterogeneous network is propagated in the subnets, between two subnets, and on the hypergraph, so as to obtain a propagation filtering signal of the heterogeneous network. Therefore, tensor characteristic data of different types of nodes of the heterogeneous network and connection of the different types of nodes are comprehensively utilized, signal transmission in a subnet, between subnets and on a hypergraph is achieved, different modes of filtering of original node signals of the heterogeneous network are achieved, theoretical basis is provided for transmission of the heterogeneous network and the original node signals on different nodes, and reliability of an algorithm is improved. Of course, it is not necessary for any product or method of practicing the invention to achieve all of the above-described advantages at the same time.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
Fig. 1 is a schematic basic flow chart of a filtering method according to an embodiment of the present invention.
Fig. 2 is a schematic flow chart of a filtering method according to an embodiment of the present invention.
Fig. 3 is a flowchart of a first implementation manner of step 207 in the filtering method according to the embodiment of the present invention.
Fig. 4 is a flowchart of a second implementation manner of step 207 in the filtering method according to the embodiment of the present invention.
Fig. 5 is a flowchart of a third implementation manner of step 207 in the filtering method according to the embodiment of the present invention.
Fig. 6 is a basic flowchart of step 306 of the filtering method according to the embodiment of the present invention.
Fig. 7 is a schematic flowchart of a filtering method according to an embodiment of the present invention.
Fig. 8 is a flowchart of a specific first implementation procedure of step 306 in the filtering method according to the embodiment of the present invention.
Fig. 9 is a flowchart of a specific second implementation procedure of step 306 in the filtering method according to the embodiment of the present invention.
Fig. 10 is a flowchart illustrating a third implementation procedure of step 306 in the filtering method according to the embodiment of the present invention.
Fig. 11 is a schematic diagram of a basic structure of a filtering apparatus according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The implementation of the invention discloses a filtering method and a filtering device, which are respectively explained in detail below.
Referring to fig. 1, fig. 1 is a flowchart of a filtering method according to an embodiment of the present invention, where the filtering method according to the embodiment of the present invention includes the following steps:
step 101, acquiring a plurality of subnets of a heterogeneous network, wherein the subnets comprise nodes of different predetermined types, and each subnet is a network composed of subnet nodes of the same predetermined type.
Typically, the different predetermined types refer to different meanings that each subnet node represents. In practical application scenarios, for example, a user posts an article in a meeting. Specifically, the predetermined type of one subnet node is a user, the predetermined type of another subnet node is a conference, and the predetermined type of another subnet node is an article. Different subnets are formed by different predetermined types of subnet nodes, and then a heterogeneous network is formed by the different subnets.
For heterogeneous networks at least comprising: one or more of nodes in the subnet, nodes between two subnets and nodes of the hypergraph. In practical applications, the nodes of the specific heterogeneous network according to the embodiment of the present invention are related to the number of node types of the node information.
Step 102, tensor characteristic data corresponding to the connection among the subnet nodes in the subnets of the subnets, the connection among two nodes between the subnets of the subnets and the connection among nodes of the hypergraph of the subnets are respectively constructed, wherein the hypergraph is formed by more than three subnets.
Here, the tensor characteristics data at least includes: one or more of an adjacency matrix within a subnet, an adjacency matrix between two subnets, and an adjacency tensor for the hypergraph. Thereby establishing tensor characteristics for different types of subnetworks.
Step 103, configuring an original node signal for the heterogeneous network.
In this step, the original node signal is a signal quantized in advance to a numerical value that can be calculated by a computer, thereby facilitating later-stage calculation. Of particular pre-quantized signalsThe method comprises the following steps: and setting node signals according to different users with different requirements. The node signal may reflect the nature of the node. The signal of each subnet is denoted as s(i)And i is 1,2, and m is the order of the heterogeneous network and takes a positive integer larger than 3.
And 104, respectively transmitting the original node signals in the subnets, between the two subnets and on the hypergraph according to the tensor feature data, and correspondingly obtaining a transmitted first filtering signal, a plurality of second filtering signals and a plurality of third filtering signals.
Step 105, determining a propagation filtered signal of the heterogeneous network according to the first filtered signal, the plurality of second filtered signals and the plurality of third filtered signals.
In the embodiment of the invention, corresponding tensor characteristic data of connection among subnet nodes in a subnet, connection between two subnet nodes and connection between hypergraph nodes are constructed by acquiring the subnet, original node signals are set for a heterogeneous network, and then the original node signals of the heterogeneous network are transmitted in the subnet, between two subnets and on the hypergraph, so that the transmission filtering signals of the heterogeneous network are obtained. Therefore, tensor characteristic data of different types of nodes of the heterogeneous network and connection of the different types of nodes are comprehensively utilized, signal transmission in a subnet, between subnets and on a hypergraph is achieved, different-mode filtering of original node signals of the heterogeneous network is achieved, the limitation of signal filtering of a complex network is reduced, and theoretical basis is provided for the transmission of the heterogeneous network and the original node signals on different nodes.
In a specific implementation process of step 103, the number m of node types is not fixed, and when the number m of node types is 1, only filtering within the subnet is required, and when the number m of node types is 2, filtering within the subnet and filtering between subnets are required. When the number m > of node types is 3, intra-subnet filtering, inter-subnet filtering, and multi-subnet filtering are required. Therefore, the following three kinds of tensor feature data are obtained specifically for the tensor feature data of the preset network corresponding to different filtering.
Filtering in the first subnetwork: constructing an adjacent matrix in the subnet corresponding to the connection of the same type of subnet nodes in the subnet; the topology inside the subnet is represented by the adjacency matrix within the subnet.
In the first process of obtaining tensor feature data, the sub-network is constructed according to the connection relationship between nodes of the node information. The nodes in the subnet inode information all belong to the same type, where i is 1,2, … m, where m represents the number of subnets. Subnet i is denoted G(i)＝{V(i),E(i)In which V is(i)Is a set of points, E(i)Is a collection of edges.
In operation, the internal topology of the sub-network is expressed by using a matrix, so that the properties of the sub-network can be researched by an algebraic method, and a computer can process the sub-network conveniently. Before representing the sub-network by using the matrix, the nodes or edges of the sub-network must be aligned in order. Here, only the case of an undirected graph is considered, and the case of a directed graph is not considered. The adjacency matrix within the subnet is a symmetric matrix.
Adjacency matrixIs the most widely used matrix. It describes the connection relationships between the various nodes and therefore contains the most basic topological properties of the network. Wherein IiRepresenting the total number of nodes that the subnet i contains,representing a complex field.
The definition of the adjacency matrix within the subnet is:
wherein, wpqFor the elements corresponding to the p-th row and q-th column of the adjacency matrix W, p and q are 1,2, …, IiAnd denotes a p-th node and a q-th node within the subnet. I isiFor the total number of nodes of the subnet i, i-1, 2.., m, the adjacency matrix of the undirected graph must be symmetric and have 0 diagonal elements. If two subnet nodes are connected, the adjacency matrix corresponding element is 1, otherwise it is 0.
Second inter-subnetwork filtering:
in the second process of obtaining the tensor feature data, the connection between the nodes between the two subnets and the connection between the information subnets is limited to the connection between every two subnets. Consider the connection relationship between subnet i and subnet j two by two, i, j ≠ 1., m i ≠ j, where m is a positive integer greater than or equal to 2. The preset network between the sub-networks is constructed according to the connection between the nodes of the two types of classes. This subnet is denoted G(i,j)＝{V(i),V(j),E(i,j)In which V is(i)Node, V, representing a subnet i(j)Representing the nodes of subnet j. E(i,j)Representing the connection relationship between the two types of nodes. I.e. the set of nodes V may be partitioned into two mutually disjoint subsets and both nodes to which each edge is attached belong to the two mutually disjoint subsets.
Adjacency matrix between subnetsIs defined as:
wherein,as a contiguous matrix W between subnetworks(i,j)Row P and column q, P being a node of the subnetwork I, P being 1,2, … Ii,IiA total number of nodes of the subnet i, j being 1., m i ≠ j, m being an integer greater than or equal to 2; q is 1,2, …, IiJ is 1,2, …, IjAnd q is a node of subnet j, IjThe total number of nodes for subnet j. Because of IiAnd IjThere is no specific size relationship, so when Ii≠IjWhen W is(i,j)Is an asymmetric matrix. If the subnet nodes between two subnets have edges connected, the corresponding element of the adjacency matrix is 1, otherwise, it is 0.
Third multi-subnet filtering:
for the third process of obtaining tensor feature data, the node connection information of the hypergraph forms an n-order hypergraph. Wherein n represents the number n ≧ 3 of subnets included in the hypergraph. The nodes connected by the n-order hypergraph form a preset network. Each n-order super edge contains n nodes, and each subnet has one and only one node contained in the n-order super edge. The hypergraph differs from the normal map in that: each edge of the latter can only connect two nodes, while the edge of the hypergraph can connect more than three nodes, so called hyperedge.
And modeling the n-order hypergraph by using n-order tensor 3 which is not less than n and not more than m. An m-th order complex network may contain multiple n-th order hypergraphs. But n must be less than or equal to m. The complex network of m order refers to a complex network with m subnets, that is, a complex network with m node types.
n-order hypergraph G(1,2,...,n)＝{V(1),V(2),...,V(n),E(1,2,...,n)}，V(1),V(2),...,V(n)Representing nodes of the n class, E(1,2,...,n)Representing an n-order superedge connecting the n-type nodes.
Adjacent tensor definition for n-order hypergraphsIs composed of
Wherein i1,i2,...,inIs the node of each sub-network, and the total number of the nodes of each sub-network is recorded as i1,i2,...,in. If nodes among the sub-networks are connected by the super edges, the corresponding element of the adjacency tensor is 1, and otherwise, the corresponding element of the adjacency tensor is 0.
Referring to fig. 2, fig. 2 is a detailed flowchart of a filtering method according to an embodiment of the present invention, where the filtering method according to the embodiment of the present invention includes the following steps:
step 201, acquiring a plurality of subnets of a heterogeneous network, wherein the subnets include nodes of different predetermined types, and each subnet is a network composed of subnet nodes of the same predetermined type.
Step 202, respectively constructing connections among the subnet nodes in the subnets of the subnets, connections between two subnet nodes of the subnets, and connections between hypergraph nodes of the subnets, and corresponding adjacency matrixes in the subnets, adjacency matrixes between the subnets, and adjacency tensors of the hypergraph, wherein the hypergraph is formed by more than three subnets.
Here, the heterogeneous network generally includes at least three types of network subnets and more.
Step 203, configuring an original node signal for the heterogeneous network.
Step 204, normalizing the tensor feature data when the tensor feature data are the adjacent matrix in the subnet and the adjacent matrix between the two subnets, and respectively propagating the original node signal once on the subnet node in the subnet and the node between the two subnets according to the normalized tensor feature data to obtain a propagated first-order first filtering signal and a plurality of first-order second filtering signals.
Step 205, calculating a projection matrix of the adjacent tensor of the hypergraph when the tensor feature data is the adjacent tensor of the hypergraph;
and 206, normalizing the projection matrix, determining a combined signal, and spreading the original node signal once on the nodes of the hypergraph according to the combined signal and the normalized projection matrix to obtain a plurality of first-order third filtering signals, wherein the combined signal is the only frequently-occurring signal or any one of a plurality of frequently-occurring signals in the hypergraph except the signal of the current subnet.
Step 207, linearly weighting the first filtered signal, the plurality of second filtered signals and the plurality of third filtered signals, and determining the propagation filtered signals of the heterogeneous network.
Step 207 specifically includes: performing linear weighting on the first-order first filtering signal, the plurality of first-order second filtering signals and the plurality of first-order third filtering signals to obtain a first-order propagation filtering signal of the heterogeneous network; and performing L iterations on the first-order first filtering signal, the first-order second filtering signals and the first-order third filtering signals, and performing linear weighting on the L-order first filtering signal, the L-order second filtering signals and the L-order third filtering signals to obtain an L-order propagation filtering signal of the heterogeneous network, wherein L is greater than or equal to 2. In particular, the propagation filtered signals are combined. Carrying out linear weighting on the first-order propagation filtering signals among the subnets, the subnets and the multiple subnets to obtain the first-order propagation filtering signals of the complex network:
whereinAnd i, j ≠ 1, 2., m i ≠ j, a(i)Is the normalized adjacency matrix for subnet i. A. the(i,j)Representing a normalized adjacency matrix between subnets i and j.The normalized projection matrix for subnet i is shown in the pth hypergraph. s(i)Representing the original node signal of subnet i.Representing the union signal of subnet i in the pth hypergraph. f. of(i)Signal s representing a subnet i(i)And (4) obtaining a signal after filtering. And performing L times of iteration on the first-order propagation filtering signal to obtain the complex network L-order propagation filtering signal.
In the embodiment of the invention, the original node signals are propagated in the heterogeneous network for establishing tensor characteristic data for one time, so that the original node signals are filtered for one time; the original node signals are propagated for many times in the heterogeneous network for establishing tensor characteristic data, so that the original node signals are filtered for many times, signal propagation in the heterogeneous network can be realized, the propagated original node signals can be filtered, a theoretical basis is provided for the heterogeneous network and the original node signals to be propagated on different nodes, and the reliability of an algorithm applied to the heterogeneous network is improved.
Generally, the specific implementation process for obtaining the L-order propagation filtered signal in step 207 includes the following three implementation manners.
The first implementation mode comprises the following steps: and when the tensor characteristic data are adjacent matrixes in the subnet, normalizing the tensor characteristic data, and respectively propagating the original node signals once on subnet nodes in the subnet according to the normalized tensor characteristic data to obtain a first-order propagated first-order filtered signal.
Referring to fig. 3, fig. 3 is a flowchart of a first implementation manner of step 207 in the filtering method according to the embodiment of the present invention. The specific first implementation manner is as follows: step 20701, obtain the adjacency matrix and degree matrix of the subnet.
Modeling from complex network tensor to obtain W(i)，Is a contiguous matrix of the network and,representing a complex field.Representing the corresponding degree matrix, the diagonal elements of the degree matrix are the degrees of the corresponding nodes, and the off-diagonal elements are zero. I isiRepresenting the total number of subnet inodes.Where k represents all nodes connected to point l.
Step 20702, normalize the adjacency matrix.
WhereinIs the adjacency matrix for the sub-network i,is a subnet i degree matrix.Is a diagonal matrix, soRepresents that D is(i)Diagonal element ofTo the power, the other elements remain unchanged.
Step 20703, first order propagation filtered signals within the subnet.
The nodes of the subnetwork i belong to the same type, where i ═ 1, 2. Defining signals on the p-th nodeAnd is replaced by adjacent nodes in a linear weighting mode.
Wherein q ∈ NpRepresenting all nodes q connected to node p,is a new signal on node p after propagation. A. the(i)Is a normalized adjacency matrix that is,is the original node signal on node q.
The first order propagation filter signals in the subnet are written in the form of vector product:
whereinA signal representing the original node of the sub-network i,representing the signal of the sub-network i after the first-order propagation of the filtered signal, A(i)Is a normalized adjacency matrix.
Step 20704, propagate the filtered signal in L order within the subnet.
And performing L times of iteration on the first-order propagation filtering signals to obtain an L-order filtering formula of the propagation filtering signals in the subnet:
h(A)＝h0I+h1A+......+hLALwhereinRepresents an identity matrix, h0,h1,...,hLRepresenting the filter coefficients and a representing the subnet normalized adjacency matrix.A signal representing the original node of the sub-network i,representing the signal of subnet i after the first order propagation filtered signal.
The second implementation mode comprises the following steps: and when the tensor characteristic data are adjacent matrixes between the two subnets, normalizing the tensor characteristic data, and respectively transmitting the original node signals once on the nodes between the two subnets according to the normalized tensor characteristic data to obtain a first-order second filtering signal.
Referring to fig. 4, fig. 4 is a flowchart of a second implementation manner of step 207 in the filtering method according to the embodiment of the present invention. The specific second implementation manner is as follows: step 20705, obtain the adjacency matrix and degree matrix between each two subnets i and j.
Obtaining W by complex network tensor modeling module(i.j)，Is the adjacency matrix between subnets i and j,representing a complex field. I isiAnd IjRepresenting the total number of nodes in subnets i and j, respectively.Representing the corresponding degree matrix, degreeThe elements of the matrix diagonal are the degrees of the corresponding nodes.
Where k represents all nodes connected to point l. The off-diagonal elements are zero.
Step 20706, normalize the adjacency matrix.
WhereinIs the adjacency matrix between subnets i and j,is a diagonal matrix, soRepresents that D is(i,j)Diagonal element ofTo the power, the other elements remain unchanged.Is a normalized adjacency matrix.
Step 20707, first order propagation of filtered signals between subnets.
Normalized adjacency matrixRepresenting the connection relationship of the subnets i and j with each other. The node signal of the sub-network i is obtained by the nodes of the connected sub-network j. Where i, j equals 1, 2.., m i ≠ j.
Between subnets i and j is a first order propagating filtered signal of the form: f. of(i)＝A(i,j)s(j)
Wherein i, j is 1, 2., m i ≠ j,is a normalized adjacency matrix that is,the original node signal representing sub-network j,representing the signal of subnet i after propagating the filtered signal.
Finding out all the subnetworks j adjacent to the subnet i to obtain a first-order propagation filtering signal between the subnetworks:
where j ∈ NiRepresenting all sub-networks j, A adjacent to sub-network i(i,j)Representing a normalized adjacency matrix between subnets i and j. s(j)Representing the original node signal of subnet j. f. of(i)Representing a propagation filtered signal obtained by a first order propagation filtered signal between subnets.
Step 20708, propagate the filtered signal in L order between subnets.
And performing L times of iteration on the first-order propagation filtering signals to obtain L-order propagation filtering signals among the subnetworks.
The third implementation mode comprises the following steps: calculating a projection matrix of the adjacent tensors of the hypergraph according to the tensor feature data when the tensor feature data are the adjacent tensors of the hypergraph; normalizing the projection matrix, determining a combined signal, and spreading the original node signal once on the nodes of the hypergraph according to the combined signal and the normalized projection matrix to obtain a plurality of first-order third filtering signals, wherein the combined signal is the only frequently-occurring signal or any one of a plurality of frequently-occurring signals in the hypergraph except the signal of the current subnet.
Wherein the only frequently occurring signal is one of the most frequently occurring signals; the plurality of frequently occurring signals are signals of frequently occurring signals, and the signals should be frequently occurring signals at the top in a frequent order.
Referring to fig. 5, fig. 5 is a flowchart illustrating a third implementation manner of step 207 in the filtering method according to the embodiment of the present invention. The third specific implementation manner is as follows: the multi-subnet here refers to the number n ≧ 3 of subnets. Analysing the signal s of the subnetwork i(i)1,2, the propagation of m on the n-order hypergraph, 3 ≦ n ≦ m. The complex network with m ≧ 3 can contain a plurality of n-order hypergraphs 3 ≦ n ≦ m.
Step 20709, calculate the adjacency tensorThe projection matrix of (2).
Tensor of order N Element (1) ofCorresponding to n-order expansion matrixThe positions of (A) are: line inColumn (i)n+1-1)In+2In+3...INI1I2...In-1+(in+2-1)In+3In+4...INI1I2...In-1+...+(iN-1)I1I2...In-1+(i1-1)I2I3...In-1+(i2-1)I3I4...In-1+...+in-1
Step 20710, normalizeThe projection matrix of (2).
WhereinIs the adjacency tensorIn the projection matrix in the i-th dimension,is W(i)Corresponding degree matrix, (D)(i))ll＝∑k(W(i))lkAnd k denotes all nodes connected to point l. The off-diagonal elements are zero.
Is thatA corresponding matrix of degrees is formed,is W(i)The transposed matrix of (2).k represents all nodes connected to point l. The off-diagonal elements are zero.
Represents that D is(i)Diagonal element ofTo the power, the other elements remain unchanged. A. the(i)Is a normalized projection matrix.
Step 20711, calculate the joint signal.
First, all hypergraphs containing subnet i are determined.
Each hypergraph corresponds to a joint signal in the ith dimension:
the i-th dimension of the joint signalSignal s with subnet i(i)Independently of the subnet signal s, other than subnet i(j)j ≠ i j ═ 1, 2.., n is relevant.In different hypergraphs, the connected subnets j are different and therefore have different values.
Secondly, the connection situation of the sub-networks except the sub-network i in the hypergraph is analyzed.
If s is(j)j ≠ i j ═ 1, 2., where n is not connected by a superedge. Otherwise s(j)As the most frequently occurring signalIf the most frequently occurring signal is not unique, then values are taken randomly from the most frequently occurring signal. Each s(j)Projection W of the value sequence of (1) and the i-dimensional adjacency tensor(i)The corresponding column order is the same. WhereinW(i)Is thatA projection matrix in the ith dimension.Is the contiguous tensor for the n-th order hypergraph.
Step 20712, first order propagation of the filtered signal for multiple subnets.
s(i)1,2, the first order propagated filtered signal of m on a hypergraph is defined as:wherein A is(i)Is a normalized projection matrix of the image data,is a joint signal of the i-th dimension, f(i)Is a sub-network signal s(i)And (4) obtaining a signal after filtering.
The subnet i can be connected with different subnets through the superedge to form different hypergraphs. Note that subnet i is contained in p different hypergraphs.
Signal s(i)1, 2.. m, the multi-subnet first-order propagating filtered signal is defined as:wherein,representing the union signal of subnet i in the pth hypergraph.The normalized projection matrix for subnet i is shown in the pth hypergraph. f. of(i)Is a sub-network signal s(i)And (4) obtaining a signal after filtering.
Step 20713, L-order propagation of filtered signals among multiple subnetworks
And performing L times of iteration on the first-order propagation filtering signal to obtain the multi-subnet L-order propagation filtering signal.
Referring to fig. 7, fig. 7 is a schematic specific flowchart of a filtering method according to an embodiment of the present invention, where the filtering method according to the embodiment of the present invention includes the following steps:
step 301, acquiring a plurality of subnets of a heterogeneous network, wherein the subnets include nodes of different predetermined types, and each subnet is a network composed of subnet nodes of the same predetermined type.
Step 302, constructing tensor feature data corresponding to the connections among the subnet nodes in the subnets of the subnets, the connections between two nodes among the subnets of the subnets, and the connections among hypergraph nodes of the subnets, respectively, wherein the hypergraph is formed by more than three subnets.
Step 303, configuring an original node signal for the heterogeneous network.
After step 303, the filtering method according to the embodiment of the present invention further includes: and converting the original node signal into a frequency domain signal according to the preset filtering mode, filtering the frequency domain signal, determining a filtering signal of the original node signal of the heterogeneous network, and further realizing frequency domain filtering of the original node signal.
And 304, according to the tensor feature data, respectively transmitting the original node signals in the subnets, between the two subnets and on the hypergraph, and correspondingly obtaining a first filtering signal, a plurality of second filtering signals and a plurality of third filtering signals after transmission.
And 305, acquiring preset filtering modes corresponding to the sub-networks, the two sub-networks and the hypergraph.
Generally, the preset filtering manner includes: one or more of a high-pass filtering mode, a low-pass filtering mode, a band-pass filtering mode and a band-stop filtering mode. Specifically, the propagated first filtered signal, the plurality of second filtered signals and the plurality of third filtered signals are filtered in the manners of high pass, low pass, band stop and the like. And transmitting the filtered signal according to the frequency domain, and obtaining a first filtered signal, a second filtered signal and a third filtered signal after transmission through inverse Fourier transform. And performing linear weighting on the obtained propagated first filtered signal, the propagated second filtered signal and the propagated third filtered signal to obtain the propagated filtered signal of the heterogeneous network.
Step 306, according to the preset filtering mode, frequency-domain filtering the first filtering signal, the plurality of second filtering signals and the plurality of third filtering signals, and determining the propagation filtering signals of the heterogeneous network.
In this step 306, the propagated first filtered signal, the propagated second filtered signal, and the propagated third filtered signal corresponding to the subnets, and the multiple subnets are filtered in the required high-pass, low-pass, band-stop, etc. manners. And then obtaining the first filtering signal, the second filtering signal and the third filtering signal after propagation through inverse Fourier transform. And performing linear weighting on the obtained propagated first filtered signal, the propagated second filtered signal and the propagated third filtered signal to obtain the propagated filtered signal of the heterogeneous network.
Referring to fig. 6, fig. 6 is a basic flowchart of step 306 of the filtering method according to the embodiment of the present invention. This step 306 specifically includes: step 3061, perform preset feature decomposition on the normalized tensor feature data. The specific implementation of step 3061 is: and when the normalized tensor feature data is an adjacent matrix in the normalized subnet, performing feature decomposition on the adjacent matrix in the normalized subnet.
And when the normalized tensor characteristic data is an adjacent matrix between the two normalized sub-networks, carrying out singular value decomposition on the adjacent matrix between the two normalized sub-networks.
And when the tensor characteristic data is an adjacent tensor, performing high-order singular value decomposition on the adjacent tensor.
Step 3062, frequency-domain filtering the first filtered signal, the plurality of second filtered signals and the plurality of third filtered signals according to the preset filtering mode and the preset feature decomposition, and correspondingly obtaining a first frequency-domain signal, a plurality of second frequency-domain signals and a plurality of third frequency-domain signals.
Step 3063, converting the first frequency domain signal, the plurality of second frequency domain signals and the plurality of third frequency domain signals to correspond to the first filtered signal, the plurality of second filtered signals and the plurality of third filtered signals, respectively.
Step 3064, linearly weighting the first filtered signal, the plurality of second filtered signals and the plurality of third filtered signals, determining a propagation filtered signal for the heterogeneous network.
In the embodiment of the invention, the original node signals are propagated in the heterogeneous network for establishing tensor characteristic data for one time, so that the original node signals are filtered for one time; or the original node signals are propagated for multiple times in the heterogeneous network for establishing the tensor characteristic data, the original node signals are filtered for multiple times, then the results of the filtering for one time or the results of the filtering for multiple times are subjected to frequency domain filtering, different filtering modes are realized, the filtering results of the filtering for the heterogeneous network are optimized, the propagated filtering signals of the heterogeneous network are determined more accurately, and the reliability of a frequency domain filtering algorithm is improved.
In this embodiment of the present invention, a specific implementation process of this step 306 is as follows.
Referring to fig. 8, fig. 8 is a flowchart of a specific first implementation procedure of step 306 in the filtering method according to the embodiment of the present invention. In this step 306, when the normalized tensor eigen data is an adjacency matrix in a normalized subnet, a specific first implementation process is performed when eigen decomposition is performed on the adjacency matrix in the normalized subnet.
Step 30601, perform a feature decomposition on the normalized adjacency matrix A of subnet i.
And (3) carrying out characteristic decomposition on the A: u Λ U ═ a-1WhereinIs a feature vector and the ith column of (1),is an orthogonal matrix. U shape-1＝UTWherein U isTRepresenting the transpose of the matrix U, U-1Representing the inverse of the matrix U. Λ is the diagonal matrix and the diagonal elements are eigenvalues of a.
Step 30602, a Fourier transform of the subnet is defined.
Fourier transform to define the subnet: f ═ U-1
Defining an inverse fourier transform of the subnet: f-1＝U
Defining the feature matrix U as a Fourier transform basis, the size of the diagonal elements of Λ reflects the height of the frequency, with larger elements corresponding to lower frequencies.
Step 30603, frequency domain filtering is performed on the signal.
Transforming the propagated first filtering signal s to the frequency domain through Fourier transform F to obtain a frequency domain response signalThe first filtered signal after propagation of r-Fs is mapped to the frequency domain as indicated
Where Ff may be viewed as the frequency domain response of the propagated first filtered signal,is the frequency domain signal of the original node signal s. Therefore h (Λ) can be seen as the system frequency domain response. F is the orthogonal transform basis of the fourier transform. Λ is the diagonal matrix, the diagonal elements are eigenvalues of a, and the off-diagonal elements are zero. The elements of the a diagonal correspond to frequencies. Larger elements correspond to lower frequencies.
By operating on r, high-pass, low-pass, band-stop, etc. filters can be designed. And keeping the signal corresponding to the high frequency, and inhibiting the low frequency signal to obtain the high-pass filtering of the signal. Otherwise, the signal corresponding to the low frequency is retained, and the high frequency signal is suppressed, so that the low pass filtering of the signal can be obtained. Similarly, band-pass and band-stop can be realized.
Step 30604, the filtered frequency domain signal is converted to a propagated first filtered signal.
s＝F-1r, whereinIs a processed frequency domain signal, and is transformed by inverse Fourier transform F-1The frequency domain signal may be converted into a propagated first filtered signal, s being the frequency domain filtered signal.
Referring to fig. 9, fig. 9 is a flowchart of a specific second implementation procedure of step 306 in the filtering method according to the embodiment of the present invention. In this step 306, when the normalized tensor feature data is an adjacency matrix between two normalized subnets, and when singular value decomposition is performed on the adjacency matrix between the two normalized subnets, a specific second implementation procedure is performed. The inter-subnet frequency domain filtering refers to frequency domain filtering between two subnets. The frequency domain filtering between the subnetworks is analyzed using the projection of the adjacency matrix,
step 30605, perform Singular Value Decomposition (SVD) on the normalized adjacency matrix.
Adjacency matrix to be normalizedi, j ≠ 1, 2., m i ≠ jProjection in the ith dimensionWherein,is A(ij)A projection matrix to the ith dimension; the projection matrix of the jth dimension is expressed as
Wherein
According to the analysis of the intra-subnet filtering, through the iteration of the L steps, the L-order filtering between subnets can be written as: h (A)u)＝h0I+h1Au+......+hL(Au)L
Au＝AATWherein I represents an identity matrix, h0,h1,...,hLRepresenting the filter coefficients and a the normalized adjacency matrix between subnets.
To AuPerforming a characteristic decomposition has the same meaning as performing a singular value decomposition of A
And performing singular value decomposition on the A to obtain: a ═ U Λ V-1
Wherein U is AuIs a, i.e. column i is auThe feature vector of (2). V isIs used to form a matrix. The diagonal elements of Λ are the singular values of a, and the off-diagonal elements are zero. The elements of the a diagonal correspond to frequencies. Larger elements correspond to lower frequencies.
Step 30606, a Fourier transform is defined.
Normalized adjacency matrix A for subnets i and j(ij)Performing SVD decomposition to obtain:
A(ij)＝UΛV-1
the fourier transform of subnet i is defined as: f ═ U-1
The inverse fourier transform is defined as: f-1U, wherein U-1Represents the inverse of the matrix U; the fourier transform of the same subnet j is defined as: f ═ V-1
The inverse fourier transform is defined as: f-1Is V, wherein V-1Represents the inverse of the matrix V; the diagonal elements of Λ are the singular values of a, and the off-diagonal elements are zero. The elements of the a diagonal correspond to frequencies. Larger elements correspond to lower frequencies.
Step 30607, frequency domain filtering is performed on the signal.
Transforming the second filtering signal s to the frequency domain through Fourier transform F between sub-networks to obtain a frequency domain response signalWherein r ═ Fs.
By operating on r, high-pass, low-pass, band-stop, etc. filters can be designed. And keeping the signal corresponding to the high frequency, and inhibiting the low frequency signal to obtain the high-pass filtering of the signal. Similarly, low-pass, band-pass, and band-stop can be implemented.
Step 30608, the filtered frequency domain signal is converted to a second filtered signal.
s＝F-1r, wherein,is a processed frequency domain signal, and is transformed by inverse Fourier transform F-1A conversion of the frequency domain signal into a second filtered signal may be obtained, s being the frequency domain filtered signal.
Referring to fig. 10, fig. 10 is a flowchart illustrating a specific third implementation procedure of step 306 in the filtering method according to the embodiment of the present invention. In this step 306, when the tensor feature data is an adjacent tensor, and the adjacent tensor is subjected to high-order singular value decomposition, a specific third implementation process is performed.
The multi-subnet here refers to the number n ≧ 3 of subnets. The multi-subnet frequency domain filtering is analyzed using a High Order Singular Value Decomposition (HOSVD) of the adjacency tensor.
Step 30609, the adjacent tensors are mapped by HOSVDDecomposition is carried out.
Using HOSVD to pair adjacent tensorsDecomposing to obtain an orthogonal matrix U of each dimension
The HOSVD represents a tensor as a multiplication of a core (core) tensor by a matrix in each dimension, where the matrix in each dimension is an orthogonal matrix.
The HOSVD decomposition is represented as:tensor thereinCore tensor Is a factor matrix of the ith dimension and can be considered as a principal component in each dimension. TensorIs a core tensor, which represents eachInteraction in dimension.Is an n-mode product operation.
A tensorAnd a matrixN-mode product ofThe elements of (a) are defined as:
step 30610, a Fourier transform is defined.
HOSVD orthogonal matrix U(i)Can pass through W(i)Performing SVD to obtain
W(i)＝U(i)Σ(i)V(i)WhereinIs the adjacency tensorA projection matrix in the ith dimension.Is a factor matrix of the ith dimension, sigma(i)Frequency information, V, representing the ith dimension(i)Representing the interaction information of the remaining dimensions. Sigma(i)Is the adjacent tensorSingular values, off-diagonal, in the i-th dimensionThe element is zero. The larger the singular value, the lower the corresponding frequency.
According to the Fourier transform analysis among the sub-networks, the ith dimension Fourier transform of the multi-sub-network is defined as:
F(i)＝(U(i))T＝(U(i))-1(ii) a The inverse fourier transform of the ith dimension is defined as: (F (i))-1U (i), whereinIs a factor matrix of the ith dimension and is an orthogonal matrix.
Step 30611, frequency domain filtering is performed on the signal.
Transforming the third filtering signal s to the frequency domain through Fourier transform F of multiple subnets to obtain a frequency domain response signalWherein r ═ Fs.
By operating on r, high-pass, low-pass, band-stop, etc. filters can be designed. And keeping the signal corresponding to the high frequency, and inhibiting the low frequency signal to obtain the high-pass filtering of the signal. Similarly, low-pass, band-pass, and band-stop can be implemented.
Step 30612, the filtered frequency domain signal is converted to a third filtered signal.
s＝F-1r wherein (a) is (b),is a processed frequency domain signal, and is transformed by inverse Fourier transform F-1A conversion of the frequency domain signal into a third filtered signal may be obtained, s being the frequency domain filtered signal.
Referring to fig. 11, fig. 11 is a schematic structural diagram of a filtering apparatus according to an embodiment of the present invention, where the filtering apparatus according to the embodiment of the present invention includes the following structures:
the obtaining module 401 is configured to obtain multiple subnets of a heterogeneous network, where the multiple subnets include nodes of different predetermined types, and each subnet is a network formed by subnet nodes of the same predetermined type.
A constructing module 402, configured to respectively construct tensor feature data corresponding to connections between the subnet nodes in the subnets of the multiple subnets, connections between two nodes between the subnets of the multiple subnets, and connections between hypergraph nodes of the multiple subnets, where the hypergraph is formed by more than three subnets.
A configuring module 403, configured to configure an original node signal for the heterogeneous network.
And the filtering signal obtaining module is used for respectively transmitting the original node signals in the subnets, between the two subnets and on the hypergraph according to the tensor characteristic data and correspondingly obtaining a first filtering signal, a plurality of second filtering signals and a plurality of third filtering signals after transmission.
A processing module 404, configured to determine a propagation filtered signal of the heterogeneous network according to the first filtered signal, the plurality of second filtered signals, and the plurality of third filtered signals.
In the embodiment of the invention, corresponding tensor characteristic data of connection among subnet nodes in a subnet, connection between two subnet nodes and connection between hypergraph nodes are constructed by acquiring the subnet, original node signals are set for a heterogeneous network, and then the original node signals of the heterogeneous network are transmitted in the subnet, between two subnets and on the hypergraph, so that the transmission filtering signals of the heterogeneous network are obtained. Therefore, tensor characteristic data of different types of nodes of the heterogeneous network and connection of the different types of nodes are comprehensively utilized, signal transmission in a subnet, between subnets and on a hypergraph is achieved, different-mode filtering of original node signals of the heterogeneous network is achieved, the limitation of signal filtering of a complex network is reduced, and theoretical basis is provided for the transmission of the heterogeneous network and the original node signals on different nodes.
It should be noted that, the apparatus according to the embodiment of the present invention is an apparatus applying the filtering method, and all embodiments of the filtering method are applicable to the apparatus and can achieve the same or similar beneficial effects.
In the filtering apparatus according to another embodiment of the present invention, the processing module 404 is specifically configured to linearly weight the first filtered signal, the plurality of second filtered signals and the plurality of third filtered signals, and determine the propagation filtered signal of the heterogeneous network.
It is noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
All the embodiments in the present specification are described in a related manner, and the same and similar parts among the embodiments may be referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for the system embodiment, since it is substantially similar to the method embodiment, the description is simple, and for the relevant points, reference may be made to the partial description of the method embodiment.
The above description is only for the preferred embodiment of the present invention, and is not intended to limit the scope of the present invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention shall fall within the protection scope of the present invention.

Claims (5)

1. A method of filtering, comprising:
acquiring a plurality of subnets of a heterogeneous network, wherein the subnets comprise nodes of different preset types, and each subnet is a network formed by subnet nodes of the same preset type;
constructing tensor characteristic data corresponding to the connection among the sub-network nodes in the sub-networks of the sub-networks, the connection among two sub-network nodes of the sub-networks and the connection among hypergraph nodes of the sub-networks respectively, wherein the hypergraph is formed by more than three sub-networks;
configuring an original node signal for the heterogeneous network;
according to the tensor characteristic data, the original node signals are respectively transmitted in the subnets, between the two subnets and on the hypergraph, and a first filtering signal, a plurality of second filtering signals and a plurality of third filtering signals after transmission are correspondingly obtained;
determining a propagation filtered signal of the heterogeneous network according to the first filtered signal, the plurality of second filtered signals and the plurality of third filtered signals, wherein determining the propagation filtered signal of the heterogeneous network according to the first filtered signal, the plurality of second filtered signals and the plurality of third filtered signals comprises:
linearly weighting the first filtered signal, the plurality of second filtered signals and the plurality of third filtered signals to determine a propagated filtered signal of the heterogeneous network; or
Acquiring preset filtering modes corresponding to the sub-networks, the two sub-networks and the hypergraph; according to the preset filtering mode, filtering the first filtering signal, the plurality of second filtering signals and the plurality of third filtering signals in a frequency domain to determine a propagation filtering signal of the heterogeneous network; or
The transmitting the original node signal in the subnet, between the two subnets and on the hypergraph respectively according to the tensor feature data, and correspondingly obtaining a first filtering signal, a plurality of second filtering signals and a plurality of third filtering signals after transmission, including:
normalizing the tensor characteristic data when the tensor characteristic data are an adjacent matrix in the subnet and an adjacent matrix between the two subnets, and respectively propagating the original node signals once on subnet nodes in the subnet and the two intersubnetwork nodes according to the normalized tensor characteristic data to obtain a first-order first filtering signal and a plurality of first-order second filtering signals after propagation; calculating a projection matrix of the adjacent tensors of the hypergraph according to the tensor feature data when the tensor feature data are the adjacent tensors of the hypergraph; normalizing the projection matrix, determining a combined signal, and spreading the original node signal once on a node of the hypergraph according to the combined signal and the normalized projection matrix to obtain a plurality of first-order third filtering signals, wherein the combined signal is a unique frequently-occurring signal or any one of a plurality of frequently-occurring signals in the hypergraph except a signal of a current subnet;
determining a propagation filtered signal of the heterogeneous network according to the first filtered signal, the plurality of second filtered signals, and the plurality of third filtered signals, comprising:
performing linear weighting on the first-order first filtering signal, the plurality of first-order second filtering signals and the plurality of first-order third filtering signals to obtain a first-order propagation filtering signal of the heterogeneous network; and performing L iterations on the first-order first filtering signal, the first-order second filtering signals and the first-order third filtering signals, and performing linear weighting on the L-order first filtering signal, the L-order second filtering signals and the L-order third filtering signals to obtain an L-order propagation filtering signal of the heterogeneous network, wherein L is greater than or equal to 2.
2. Filtering method according to claim 1,
the frequency-domain filtering the first filtered signal, the plurality of second filtered signals and the plurality of third filtered signals according to the preset filtering mode to determine the propagation filtered signals of the heterogeneous network includes:
performing preset feature decomposition on the normalized tensor feature data;
decomposing according to the preset filtering mode and the preset characteristics, filtering the first filtering signal, the plurality of second filtering signals and the plurality of third filtering signals in a frequency domain, and correspondingly obtaining a first frequency domain signal, a plurality of second frequency domain signals and a plurality of third frequency domain signals;
converting the first frequency domain signal, the plurality of second frequency domain signals and the plurality of third frequency domain signals to correspond to the first filtered signal, the plurality of second filtered signals and the plurality of third filtered signals, respectively;
linearly weighting the first filtered signal, the plurality of second filtered signals, and the plurality of third filtered signals to determine a propagated filtered signal for the heterogeneous network.
3. Filtering method according to claim 2,
the preset feature decomposition is carried out on the normalized tensor feature data, and the method comprises the following steps:
when the normalized tensor feature data is an adjacent matrix in a normalized subnet, performing feature decomposition on the adjacent matrix in the normalized subnet;
when the normalized tensor characteristic data is an adjacent matrix between two normalized sub-networks, carrying out singular value decomposition on the adjacent matrix between the two normalized sub-networks;
and when the tensor characteristic data is an adjacent tensor, performing high-order singular value decomposition on the adjacent tensor.
4. Filtering method according to claim 1,
the constructing tensor feature data corresponding to the connections among the subnet nodes in the subnets of the plurality of subnets, the connections between two nodes among the subnets of the plurality of subnets, and the connections among the hypergraph nodes of the plurality of subnets, respectively, includes:
and respectively constructing the connection among the subnet nodes in the subnets of the subnets, the connection among two nodes among the subnets of the subnets, the connection among the hypergraph nodes of the subnets, the corresponding adjacency matrix in the subnets, the adjacency matrix among the subnets and the adjacency tensor of the hypergraph.
5. A filtering apparatus, comprising:
the device comprises an acquisition module, a processing module and a processing module, wherein the acquisition module is used for acquiring a plurality of subnets of a heterogeneous network, the subnets comprise nodes of different preset types, and each subnet is a network formed by subnet nodes of the same preset type;
the constructing module is used for respectively constructing tensor characteristic data corresponding to the connection among the subnet nodes in the subnets of the subnets, the connection among two nodes among the subnets of the subnets and the connection among nodes of the hypergraph of the subnets, wherein the hypergraph is formed by more than three subnets;
the configuration module is used for configuring an original node signal for the heterogeneous network;
a filtered signal obtaining module, configured to separately propagate the original node signal in the subnet, between the two subnets, and on the hypergraph according to the tensor feature data, and correspondingly obtain a propagated first filtered signal, a plurality of second filtered signals, and a plurality of third filtered signals;
a processing module, configured to determine a propagation filtered signal of the heterogeneous network according to the first filtered signal, the plurality of second filtered signals, and the plurality of third filtered signals, where the processing module is specifically configured to linearly weight the first filtered signal, the plurality of second filtered signals, and the plurality of third filtered signals, and determine the propagation filtered signal of the heterogeneous network; or
Acquiring preset filtering modes corresponding to the sub-networks, the two sub-networks and the hypergraph; according to the preset filtering mode, filtering the first filtering signal, the plurality of second filtering signals and the plurality of third filtering signals in a frequency domain to determine a propagation filtering signal of the heterogeneous network; or
The transmitting the original node signal in the subnet, between the two subnets and on the hypergraph respectively according to the tensor feature data, and correspondingly obtaining a first filtering signal, a plurality of second filtering signals and a plurality of third filtering signals after transmission, including:
normalizing the tensor characteristic data when the tensor characteristic data are an adjacent matrix in the subnet and an adjacent matrix between the two subnets, and respectively propagating the original node signals once on subnet nodes in the subnet and the two intersubnetwork nodes according to the normalized tensor characteristic data to obtain a first-order first filtering signal and a plurality of first-order second filtering signals after propagation; calculating a projection matrix of the adjacent tensors of the hypergraph according to the tensor feature data when the tensor feature data are the adjacent tensors of the hypergraph; normalizing the projection matrix, determining a combined signal, and spreading the original node signal once on a node of the hypergraph according to the combined signal and the normalized projection matrix to obtain a plurality of first-order third filtering signals, wherein the combined signal is a unique frequently-occurring signal or any one of a plurality of frequently-occurring signals in the hypergraph except a signal of a current subnet;
determining a propagation filtered signal of the heterogeneous network according to the first filtered signal, the plurality of second filtered signals, and the plurality of third filtered signals, comprising:
performing linear weighting on the first-order first filtering signal, the plurality of first-order second filtering signals and the plurality of first-order third filtering signals to obtain a first-order propagation filtering signal of the heterogeneous network; and performing L iterations on the first-order first filtering signal, the first-order second filtering signals and the first-order third filtering signals, and performing linear weighting on the L-order first filtering signal, the L-order second filtering signals and the L-order third filtering signals to obtain an L-order propagation filtering signal of the heterogeneous network, wherein L is greater than or equal to 2.
CN201610554543.7A 2016-07-14 2016-07-14 A kind of filtering method and device Active CN106209022B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610554543.7A CN106209022B (en) 2016-07-14 2016-07-14 A kind of filtering method and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610554543.7A CN106209022B (en) 2016-07-14 2016-07-14 A kind of filtering method and device

Publications (2)

Publication Number Publication Date
CN106209022A CN106209022A (en) 2016-12-07
CN106209022B true CN106209022B (en) 2019-03-26

Family

ID=57475237

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610554543.7A Active CN106209022B (en) 2016-07-14 2016-07-14 A kind of filtering method and device

Country Status (1)

Country Link
CN (1) CN106209022B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107545151B (en) * 2017-09-01 2020-10-16 中南大学 Drug relocation method based on low-rank matrix filling

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101286226A (en) * 2008-05-23 2008-10-15 杭州华三通信技术有限公司 Diminished image digital filtering method and device
EP2296272A2 (en) * 2009-09-14 2011-03-16 Soonhan Engineering Corp. Method for detecting and filtering a random noise signal

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101286226A (en) * 2008-05-23 2008-10-15 杭州华三通信技术有限公司 Diminished image digital filtering method and device
EP2296272A2 (en) * 2009-09-14 2011-03-16 Soonhan Engineering Corp. Method for detecting and filtering a random noise signal

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
基于有限脉冲反应和径向基神经网络的触电信号识别;关海鸥等;《农业工程学报》;20130430;第29卷(第8期);第187-194页

Also Published As

Publication number Publication date
CN106209022A (en) 2016-12-07

Similar Documents

Publication Publication Date Title
Zhang et al. Introducing hypergraph signal processing: Theoretical foundation and practical applications
Perraudin et al. Global and local uncertainty principles for signals on graphs
Shomorony et al. Sampling large data on graphs
Blayo et al. Data analysis: How to compare Kohonen neural networks to other techniques?
Ramakrishna et al. A user guide to low-pass graph signal processing and its applications: Tools and applications
Lapidus et al. Dirac operators and geodesic metric on the harmonic Sierpinski gasket and other fractal sets
Lu et al. Fast graph Fourier transforms based on graph symmetry and bipartition
CN110993037A (en) Protein activity prediction device based on multi-view classification model
Quqa et al. Seismic structural health monitoring using the modal assurance distribution
Ribeiro et al. Graph signal processing in a nutshell
Songsiri Sparse autoregressive model estimation for learning Granger causality in time series
Slavakis et al. Online dictionary learning from big data using accelerated stochastic approximation algorithms
Pena-Pena et al. T-HGSP: Hypergraph signal processing using t-product tensor decompositions
Sondak et al. Learning a reduced basis of dynamical systems using an autoencoder
Vasan et al. Experimental validation of LS-SVM based fault identification in analog circuits using frequency features
CN106209022B (en) A kind of filtering method and device
Xiao et al. Joint sampling and reconstruction of time-varying signals over directed graphs
Ouyang et al. Fast principal component extraction by a weighted information criterion
CN116992334A (en) Academic-oriented network node classification method and device
Liu et al. Graph signal coarsening: Dimensionality reduction in irregular domain
Ding Spectral analysis of large block random matrices with rectangular blocks
Matei et al. Distributed algorithms for optimization problems with equality constraints
Sardellitti et al. Topological Signal Processing Over Generalized Cell Complexes
Yankelevsky et al. Dictionary learning for high dimensional graph signals
Tseng et al. A rational graph filter method for GFT centrality computation

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant