WO2023065525A1 - 对象特征矩阵确定方法、装置、设备和存储介质 - Google Patents
对象特征矩阵确定方法、装置、设备和存储介质 Download PDFInfo
- Publication number
- WO2023065525A1 WO2023065525A1 PCT/CN2021/140980 CN2021140980W WO2023065525A1 WO 2023065525 A1 WO2023065525 A1 WO 2023065525A1 CN 2021140980 W CN2021140980 W CN 2021140980W WO 2023065525 A1 WO2023065525 A1 WO 2023065525A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- matrix
- function
- heterogeneous
- feature
- network
- Prior art date
Links
- 239000011159 matrix material Substances 0.000 title claims abstract description 571
- 238000000034 method Methods 0.000 title claims abstract description 85
- 230000006870 function Effects 0.000 claims description 270
- 238000005314 correlation function Methods 0.000 claims description 78
- 230000015654 memory Effects 0.000 claims description 16
- 238000010276 construction Methods 0.000 claims description 14
- 230000008569 process Effects 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 4
- 238000006243 chemical reaction Methods 0.000 abstract description 8
- 238000000354 decomposition reaction Methods 0.000 description 9
- 238000010586 diagram Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 2
- 238000013500 data storage Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 230000006855 networking Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000013401 experimental design Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000010295 mobile communication Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/55—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/01—Social networking
Definitions
- the present disclosure relates to a method, device, equipment and storage medium for determining an object feature matrix.
- Efficient classification of objects such as images, biomolecules, social networking site users, etc. relies on building effective image classifiers.
- the premise of constructing an effective object classifier is to extract an object feature matrix that effectively characterizes the features of the sample object.
- a method, device, device, and storage medium for determining an object feature matrix are provided.
- an embodiment of the present disclosure provides a method for determining an object feature matrix, including:
- Object feature matrices of various types of objects in the object dataset are reconstructed by using the low-rank feature matrix.
- the method also includes:
- a minimum objective function including:
- the minimization objective function is constructed based on the heterogeneous feature correlation function, the attribute heterogeneous network function, the first constraint function and the second constraint function.
- the method also includes:
- the minimization objective function is constructed based on the heterogeneous feature correlation function, the attribute heterogeneous network function, the first constraint function, the second constraint function and the unknown noise correlation constraint function.
- the determining low-rank feature matrices of various types of objects based on the minimized objective function includes:
- the alternate direction multiplier method is used to solve the minimized objective function to determine the low-rank feature matrices of various types of objects.
- the method of using alternating direction multipliers to solve the minimized objective function includes:
- alternating direction multiplier method to alternately solve the first weight matrix in the minimized objective function, the second weight matrix, the low-rank feature matrix, the heterogeneous association network matrix, and the multi-layer property heterogeneity network matrix base matrix until reaching the preset number of iterations or convergence.
- the basis matrix of the network matrix including:
- the object data set is an image data set, and objects in the object data set are images.
- an apparatus for determining an object feature matrix including:
- a matrix acquisition unit configured to acquire heterogeneous feature correlation matrices and corresponding correlation indicator matrices of various types of objects in the object dataset, as well as multi-layer attribute heterogeneous network matrices of various types of objects;
- a function construction unit configured to construct a heterogeneous feature correlation function based on the heterogeneous feature correlation matrix and the corresponding correlation indicator matrix; construct an attribute heterogeneous network function based on the multi-layer attribute heterogeneous network matrix; and, based on The heterogeneous feature correlation function and the attribute heterogeneous network function construct a minimum objective function;
- a solving unit configured to calculate low-rank feature matrices of various types of objects based on the minimized objective function
- the object feature matrix determining unit is configured to use the low-rank feature matrix to reconstruct object feature matrices of various types of objects in the object dataset.
- the function construction unit is further configured to construct a first constraint function based on a first weight matrix in the heterogeneous feature correlation function; construct a second constraint based on a second weight matrix in the attribute heterogeneous network function function; constructing the minimization objective function based on the heterogeneous feature correlation function, the attribute heterogeneous network function, the first constraint function and the second constraint function.
- the function construction unit is further configured to construct an unknown noise correlation constraint function; based on the heterogeneous feature correlation function, the attribute heterogeneous network function, the first constraint function, the second constraint function and The unknown noise correlation constraint function is used to construct the minimization objective function.
- the solving unit is specifically configured to use an alternating direction multiplier method to solve the minimized objective function to determine low-rank feature matrices of various types of objects.
- the solving unit is specifically configured to sequentially and alternately solve the first weight matrix, the second weight matrix, the low-rank feature matrix, and the heterogeneous association network matrix in the minimized objective function by using the alternating direction multiplier method , and the basis matrix of each multi-layer attribute heterogeneous network matrix, until reaching the preset number of iterations or convergence.
- the solving unit is specifically configured to set the first weight matrix, the second weight matrix, the low-rank feature matrix, and the heterogeneous correlation network matrix to fixed values, and for the multi-layer attribute
- the base matrix of the heterogeneous network matrix seeks the first partial derivative, and when the first partial derivative value is zero, determine the base matrix of the multi-layer attribute heterogeneous network matrix; set the first weight matrix, the second The basic matrix of the two-weight matrix, the low-rank feature matrix and the multi-layer attribute heterogeneous network matrix is a fixed value, and the second partial derivative is calculated for the heterogeneous correlation network matrix, and when the second partial derivative value is zero.
- determine the heterogeneous relational network matrix set the base matrix of the first weight matrix, the second weight matrix, the heterogeneous relational network matrix and the multi-layer property heterogeneous network matrix to a fixed value, for the Find the third partial derivative of the low-rank feature matrix, and determine the low-rank feature matrix when the third partial derivative is zero; set the first weight matrix, the low-rank feature matrix, and the heterogeneous correlation
- the network matrix and the base matrix of the multi-layer attribute heterogeneous network matrix are fixed values, the fourth partial derivative is calculated for the second weight matrix, and the second weight is determined when the fourth partial derivative value is zero
- Matrix set the base matrix of the second weight matrix, the low-rank feature matrix, the heterogeneous association network matrix and the multi-layer attribute heterogeneous network matrix to a fixed value, and find the second weight matrix for the first weight matrix five partial derivatives, determining the first weight matrix when the fifth partial derivative is zero.
- the object data set is an image data set, and objects in the object data set are images.
- the matrix acquisition unit is specifically configured to acquire raw data of various types of objects in the object data set; process the raw data of various types of objects in the object data set to obtain processed data; based on the For the processed data, heterogeneous feature correlation matrices and corresponding correlation indicator matrices of various types of objects in the object dataset, as well as multi-layer attribute heterogeneous network matrices of various types of objects are obtained.
- the function construction unit is specifically configured to add the heterogeneous feature correlation function and the attribute heterogeneous network function to obtain the sum of the heterogeneous feature correlation function and the attribute heterogeneous network function ; Construct a minimum objective function according to the sum of the heterogeneous feature correlation function and the attribute heterogeneous network function.
- An electronic device comprising: a memory and one or more processors, wherein computer-readable instructions are stored in the memory; when the one or more processors execute the computer-readable instructions, the one or more A processor executes the steps of the method for determining an object feature matrix provided in any one embodiment of the present disclosure.
- One or more non-volatile computer-readable storage media storing computer-readable instructions.
- the computer-readable instructions are executed by one or more processors, the one or more processors execute any one of the embodiments of the present disclosure.
- the steps of the method for determining the object characteristic matrix are provided.
- FIG. 1 is a flowchart of a method for determining an object feature matrix provided by one or more embodiments of the present disclosure
- Fig. 2 is a flow chart of a method for determining an object feature matrix provided by one or more embodiments of the present disclosure
- Fig. 3 is a flow chart of a method for determining an object feature matrix provided by one or more embodiments of the present disclosure
- Fig. 4 is a schematic structural diagram of an object determination device provided by one or more embodiments of the present disclosure.
- Fig. 5 is a schematic structural diagram of an electronic device provided by one or more embodiments of the present disclosure.
- An embodiment of the present disclosure provides a method for determining an object feature matrix, which is used to solve the problem of information loss caused by homogeneous data conversion during the fusion (that is, matrix decomposition) process of existing methods for determining object features based on matrix decomposition, and the known heterogeneous It solves the problem of incomplete qualitative data association, and then more accurately characterizes the characteristics of objects in object datasets.
- the objects in the embodiments of the present disclosure may be objects of types such as images, biomolecules, and users of social networking sites.
- the method for determining the object feature matrix in the present disclosure is executed by the electronic device or an application program in the electronic device.
- the electronic device may be a device such as a tablet computer, a mobile phone, a notebook computer, or a server, and the present disclosure does not impose any limitation on the specific type of the electronic device.
- the present disclosure does not limit the type of the operating system of the electronic device. For example, Android system, Linux system, Windows system, iOS system, etc.
- Fig. 1 is a flowchart of a method for determining an object feature matrix provided by an embodiment of the present disclosure. As shown in FIG. 1 , the method for determining an object feature matrix provided by an embodiment of the present disclosure includes steps S101-S106.
- Step S101 Obtain the heterogeneous feature correlation matrix and the corresponding correlation indicator matrix of various types of objects in the object dataset, as well as the multi-layer attribute heterogeneous network matrix of various types of objects.
- the electronic device can process the original data of various types of objects in the object data set, and obtain the heterogeneous feature correlation matrix, the corresponding correlation relationship indicator matrix and the multi-layer attribute heterogeneous network matrix of various types of objects,
- the aforementioned matrix can also be obtained by reading pre-stored data in the storage.
- the embodiment of the present disclosure adopts Represents the heterogeneous feature correlation matrix between n i samples of the i-th type of object, and uses H ij to represent the correlation indicator matrix corresponding to R ij and having the same dimension as R ij .
- the embodiment of the present disclosure adopts t ⁇ 1,2,,t i ⁇ represents the multi-attribute heterogeneous network matrix of the i-th type object, where t i represents the attribute heterogeneous network collected from the i-th type object to t i sources, d it Indicates that the t-th attribute heterogeneous network has d it attributes.
- Step S102 constructing a heterogeneous feature correlation function based on the heterogeneous feature correlation matrix and the corresponding correlation indicator matrix.
- the heterogeneous feature correlation function also includes the first weight matrix, the low-rank feature matrix corresponding to various types of objects, and the heterogeneous feature matrix between various types of objects. Quality association network matrix.
- the heterogeneous feature correlation function is expressed by formula 1.1.
- e represents Hadamard product
- k i and k j respectively represent the dimensions of the i type or j type low-rank feature matrix;
- the first weight matrix assigned by a heterogeneous association network, for It can be approximated as the reconstruction loss and is used to distinguish the observed relationship from the unobserved relationship, and make the observed relationship maintained in the reconstructed R ij .
- Step S103 Construct an attribute heterogeneous network function based on the multi-layer attribute heterogeneous network matrix.
- the attribute heterogeneous network function includes, in addition to the aforementioned multiple attribute heterogeneous network matrices, the second weight matrix, the low-rank feature matrix corresponding to various types of objects, and the attribute heterogeneity of various types of objects Basis matrix after network decomposition.
- the attribute heterogeneous network function is represented by Formula 1.2.
- attribute heterogeneous data can be directly decomposed by using the attribute heterogeneous network function to avoid information loss caused by homogeneous conversion.
- Step S104 Based on the heterogeneous feature correlation function and attribute heterogeneous network function, construct the minimization objective function.
- constructing the minimization objective function based on the heterogeneous feature correlation function and the attribute heterogeneous network function may be the addition of the heterogeneous feature correlation function and the attribute heterogeneous network function.
- the obtained minimization objective function is expressed by formula 1.3
- Step S105 Calculate low-rank feature matrices of various types of objects based on the minimized objective function.
- the calculation of the low-rank feature matrices of various types of objects based on the minimization objective function is to minimize the first weight matrix in the objective function, the low-rank feature matrices corresponding to various types of objects, and the The heterogeneous association network matrix among them, the second weight matrix, and the base matrix after the attribute heterogeneous network decomposition of various types of objects are decomposed, and the low-rank feature matrix G of various types of objects is obtained in the process of collaborative decomposition.
- the minimization objective function is non-convex on G, S, U, ⁇ r , ⁇ h , so the Alternating Direction Method of Multipliers (Alternating Direction Method of Multipliers , ADMM) to optimize and solve it.
- the four parameters of G, S, U, ⁇ r , ⁇ h can be set as constants, and the other one can be optimized at the same time, and iterated repeatedly until all parameters are solved.
- the calculation of G, S, U, ⁇ r , ⁇ h using the method of alternating direction multipliers is described below.
- Step S106 Reconstruct object feature matrices of various types of objects in the object dataset using low-rank feature matrices.
- the low-rank feature matrix After determining the low-rank feature matrix of various types of objects, the low-rank feature matrix can be used to reconstruct the object feature matrix. In some embodiments of the present disclosure, after the low-rank feature matrices of various types of objects are determined, the low-rank feature matrices may be directly used as object feature matrices of corresponding types of objects.
- the object feature matrix determination method constructs the heterogeneous feature correlation function through the heterogeneous feature correlation matrix and the corresponding correlation relationship indicator matrix, so as to realize the heterogeneity correlation and potential heterogeneity of the heterogeneity using the correlation relationship indicator matrix Associations are distinguished.
- the attribute heterogeneous network function is constructed by using the low-rank feature matrix corresponding to various types of objects and the decomposed basis matrix of the attribute heterogeneous network of various types of objects, and the minimum
- the objective function is to minimize the objective function to the first weight matrix, low-rank feature matrix, heterogeneous association network matrix, second weight matrix and attribute heterogeneity of various types of objects in the heterogeneous feature correlation function and attribute heterogeneous network function.
- the base matrices after qualitative network decomposition are collaboratively classified to obtain the low-rank feature matrices of various types of objects, and the object feature matrices of various types of objects are determined based on the low-rank feature matrices of various types of objects.
- the topology of the network and the attribute information of the nodes can be fused together, so as to make up for the cold start problem caused by the lack of known associations, and further reduce the information caused by the homogeneous conversion of the attribute heterogeneous network. loss.
- Fig. 2 is a flow chart of a method for determining an object feature matrix provided by an embodiment of the present disclosure. As shown in FIG. 2 , the method provided by the embodiment of the present disclosure includes steps S201-S207.
- Step S201 Obtain the heterogeneous feature correlation matrix and the corresponding correlation indicator matrix of various types of objects in the object dataset, as well as the multi-layer attribute heterogeneous network matrix of various types of objects.
- Step S202 constructing a heterogeneous feature correlation function based on the heterogeneous feature correlation matrix and the corresponding correlation indicator matrix.
- Step S203 Construct an attribute heterogeneous network function based on the multi-layer attribute heterogeneous network matrix.
- Step S204 constructing a first constraint function based on the first weight matrix in the heterogeneous feature correlation function; and constructing a second constraint function based on the second weight matrix in the attribute heterogeneous network function.
- Step S205 Based on the heterogeneous feature correlation function, the attribute heterogeneous network function, the first constraint function and the second constraint function, construct the minimization objective function.
- the first constraint function is first constructed based on the first weight matrix in the heterogeneous feature correlation function, and the second constraint function is constructed based on the second weight matrix in the attribute heterogeneous network function .
- both the first constraint function and the second constraint function are regular terms based on the l2 norm.
- the first constraint function is
- the second constraint function is
- the aforementioned vec( ⁇ r ) is a vector obtained by stacking and splicing rows of ⁇ r
- vec( ⁇ h ) is a vector obtained by stacking and splicing rows of ⁇ h .
- ⁇ >0, ⁇ >0 are used to control the complexity of vec( ⁇ r ) and vec( ⁇ h ).
- ⁇ , ⁇ can also help to selectively integrate different heterogeneous relational data sources and attribute heterogeneous data sources.
- the minimization objective function obtained in the embodiment of the present disclosure is expressed by formula 1.4.
- Step S206 Calculate low-rank feature matrices of various types of objects based on the minimized objective function.
- Step S207 Using the low-rank feature matrix to reconstruct object feature matrices of various types of objects in the object dataset.
- the object characteristic matrix determination method constructs the first constraint function based on the first weight matrix in the heterogeneous feature correlation function, and constructs the second constraint function based on the second weight matrix in the attribute heterogeneous network function, through
- the first constraint function, the second constraint function, the heterogeneous feature correlation function, and the attribute heterogeneous network function construct the minimization objective function, and calculate the characteristics of each object based on the minimization objective function, making up for the use of only a single heterogeneous correlation matrix And a single homogenous incidence matrix may not give reliable predictions of defects.
- Fig. 3 is a flowchart of a method for determining an object provided by an embodiment of the present disclosure. As shown in FIG. 3 , the method provided by the embodiment of the present disclosure includes steps S301-S308.
- Step S301 Obtain the heterogeneous feature correlation matrix and the corresponding correlation indicator matrix of various types of objects in the object dataset, as well as the multi-layer attribute heterogeneous network matrix of various types of objects.
- Step S302 Construct a heterogeneous feature correlation function based on the heterogeneous feature correlation matrix and the corresponding correlation indicator matrix.
- Step S303 Construct an attribute heterogeneous network function based on the multi-layer attribute heterogeneous network matrix.
- Step S304 constructing a first constraint function based on the first weight matrix in the heterogeneous feature correlation function; and constructing a second constraint function based on the second weight matrix in the attribute heterogeneous network function.
- Step S305 constructing an unknown noise correlation constraint function.
- the association relationship indicator matrix H ij distinguishes the observed association relationship from the unobserved association relationship, but only constrains the observed association relationship, so it is likely to leave A large number of unobserved associations are left unconstrained, which in turn leads to a large amount of noise in the target incidence matrix.
- an unknown noise correlation constraint function is also constructed, and a minimization objective function is constructed based on the unknown noise correlation constraint function.
- the unknown noise correlation constraint function is in is a constraint on the unknown noise association, and ⁇ is used to control the complexity of this term.
- Step S306 Based on the heterogeneous feature correlation function, the attribute heterogeneous network function, the first constraint function, the second constraint function and the unknown noise correlation constraint function, construct the minimization objective function.
- the minimization objective function constructed in the embodiment of the present disclosure is
- Step S307 Calculate low-rank feature matrices of various types of objects based on the minimized objective function.
- Step S308 Reconstruct object feature matrices of various types of objects in the object dataset using low-rank feature matrices.
- the alternate direction multiplier method can be used to solve the minimization objective function and determine the low-rank feature matrix of various types of objects. Specifically, using the alternating direction multiplier method to calculate the low-rank feature matrix of various types of objects, in order to use the alternating direction multiplier method to alternately solve the first weight matrix, the second weight matrix, and the low-rank feature matrix in the minimized objective function , the heterogeneous association network matrix, and the basis matrix of each multi-layer attribute heterogeneous network matrix, until a preset number of iterations or convergence is reached.
- solving the minimization objective function by using the method of alternating direction multipliers may include steps S401-S405.
- the following uses solving formula 1.5 as an example to analyze the alternating direction multiplier algorithm adopted in the embodiment of the present disclosure.
- Equation 1.5 Before calculating Equation 1.5, first introduce the Lagrange multiplier that constrains G i ⁇ 0 So Equation 1.5 can be equivalent to Equation 1.6.
- Steps S401-S405 may then be performed using an alternating direction multiplier algorithm based on Formula 1.6.
- Step S401 Set the first weight matrix, the second weight matrix, the low-rank feature matrix and the heterogeneous correlation network matrix to fixed values, and calculate the first partial derivative for the basis matrix of the multi-layer attribute heterogeneous network matrix, and in the first partial derivative When the value is zero, determines the basis matrix of the multi-layer attribute heterogeneous network matrix.
- the base matrix of the multi-layer attribute heterogeneous network matrix is
- Step S402 Set the base matrix of the first weight matrix, the second weight matrix, the low-rank feature matrix, and the multi-layer attribute heterogeneous network matrix to fixed values, calculate the second partial derivative for the heterogeneous correlation network matrix, and in the second partial derivative A value of zero determines the heterogeneous association network matrix.
- Step S403 Set the base matrices of the first weight matrix, the second weight matrix, the heterogeneous association network matrix and the multi-layer attribute heterogeneous network matrix to fixed values, calculate the third partial derivative for the low-rank feature matrix, and in the third partial derivative A value of zero determines the low-rank eigenmatrix.
- the polynomial factor ⁇ i can be obtained by making Obtained from the Karush-Kuhn-Tucker (KKT) condition:
- Equations 1.11, 1.12 and 1.13 can be defined as and Therefore, the low-rank feature matrix G can be updated as:
- Step S404 Set the base matrix of the first weight matrix, low-rank feature matrix, heterogeneous association network matrix and multi-layer attribute heterogeneous network matrix to fixed values, calculate the fourth partial derivative for the second weight matrix, and obtain the fourth partial derivative in the fourth partial derivative When the value is zero, the second weight matrix is determined.
- the solution to formula 1.15 can be regarded as a quadratic programming problem about vec( ⁇ h ), which can be solved by introducing Lagrangian multipliers based on Selective Non-Matrix Factorization (SNMF) algorithm.
- SNMF Selective Non-Matrix Factorization
- Step S405 Set the base matrix of the second weight matrix, low-rank feature matrix, heterogeneous association network matrix, and multi-layer attribute heterogeneous network matrix to fixed values, calculate the fifth partial derivative for the first weight matrix, and obtain the fifth partial derivative in the fifth partial derivative When the value is zero, the first weight matrix is determined.
- Equation 1.16 can be simplified as:
- Equation 1.17 can be regarded as a quadratic programming problem about vec( ⁇ r ), and it can also be solved by introducing Lagrangian multipliers based on the algorithm of selective matrix decomposition. It should be noted that, in the specific embodiment of the aforementioned steps S401-S405, the sequence is not limited.
- Fig. 4 is a schematic structural diagram of an object determining device provided by an embodiment of the present disclosure.
- the device for determining an object provided by the embodiment of the present disclosure includes a matrix acquiring unit 401 , a function building unit 402 , a solving unit 403 and an object characteristic matrix determining unit 404 .
- the matrix acquiring unit 401 is configured to acquire heterogeneous feature correlation matrices and corresponding correlation indicator matrices of various types of objects in the object dataset, as well as multi-layer attribute heterogeneous network matrices of various types of objects;
- the function construction unit 402 is configured to construct a heterogeneous feature correlation function based on the heterogeneous feature correlation matrix and the corresponding correlation indicator matrix; construct an attribute heterogeneous network function based on the multi-layer attribute heterogeneous network matrix; and, based on the heterogeneous feature correlation function and Attribute heterogeneous network function, construct the minimization objective function;
- the solution unit 403 is configured to calculate the low-rank feature matrix of various types of objects based on the minimization of the objective function
- the object feature matrix determining unit 404 is configured to reconstruct object feature matrices of various types of objects in the object dataset using low-rank feature matrices.
- the heterogeneous feature correlation function is constructed through the heterogeneous feature correlation matrix and the corresponding correlation relationship indicator matrix, so as to realize the use of the correlation relationship indicator matrix for heterogeneous heterogeneous correlation and potential differentiate heterogeneous associations.
- the attribute heterogeneous network function is constructed by using the low-rank feature matrix corresponding to various types of objects and the decomposed basis matrix of the attribute heterogeneous network of various types of objects, and the minimum
- the objective function is to minimize the objective function to the first weight matrix, low-rank feature matrix, heterogeneous association network matrix, second weight matrix and attribute heterogeneity of various types of objects in the heterogeneous feature correlation function and attribute heterogeneous network function.
- the base matrices after qualitative network decomposition are collaboratively classified to obtain the low-rank feature matrices of various types of objects, and the object feature matrices of various types of objects are determined based on the low-rank feature matrices of various types of objects.
- the topology of the network and the attribute information of the nodes can be fused together, thereby making up for the cold start problem caused by the lack of known associations, and further reducing the information caused by the homogeneous conversion of the attribute heterogeneous network. loss.
- the function construction unit 402 is further configured to construct the first constraint function based on the first weight matrix in the heterogeneous feature correlation function, and construct the second constraint function based on the second weight matrix in the attribute heterogeneous network function function.
- the function construction unit 402 constructs the minimization objective function based on the heterogeneous feature correlation function, the attribute heterogeneous network function, the first constraint function and the second constraint function.
- the function construction unit 402 is further configured to construct an unknown noise correlation constraint function.
- the function construction unit 402 constructs the minimization objective function based on the heterogeneous feature correlation function, the attribute heterogeneity network function, the first constraint function, the second constraint function and the unknown noise correlation constraint function.
- the solving unit 403 uses an alternating direction multiplier method to solve the minimization objective function, and determine low-rank feature matrices of various types of objects.
- the solving unit 403 uses the alternating direction multiplier method to alternately solve the first weight matrix, the second weight matrix, the low-rank feature matrix, and the heterogeneous association network matrix in the minimized objective function. , and the basis matrix of each multi-layer attribute heterogeneous network matrix, until reaching the preset number of iterations or convergence.
- the solving unit 403 calculates the first partial derivative of the base matrix of the multi-layer attribute heterogeneous network matrix by setting the first weight matrix, the second weight matrix, the low-rank feature matrix and the heterogeneous correlation network matrix as fixed values, When the first partial derivative value is zero, determine the base matrix of the multi-layer attribute heterogeneous network matrix;
- the solving unit 403 calculates the second partial derivative of the heterogeneous association network matrix by setting the first weight matrix, the second weight matrix, the low-rank feature matrix and the base matrix of the multi-layer attribute heterogeneous network matrix as fixed values, When the second partial derivative value is zero, determine the heterogeneous association network matrix;
- the solving unit 403 calculates the third partial derivative for the low-rank feature matrix by setting the base matrix of the first weight matrix, the second weight matrix, the heterogeneous association network matrix and the multi-layer property heterogeneous network matrix as fixed values, When the third partial derivative value is zero, determine the low-rank characteristic matrix;
- the solving unit 403 calculates the fourth partial derivative for the second weight matrix by setting the base matrix of the first weight matrix, low-rank feature matrix, heterogeneous association network matrix, and multi-layer property heterogeneous network matrix as fixed values, When the fourth partial derivative value is zero, determine the second weight matrix;
- the solving unit 403 calculates the fifth partial derivative of the first weight matrix by setting the base matrix of the second weight matrix, low-rank feature matrix, heterogeneous association network matrix and multi-layer property heterogeneous network matrix as fixed values, When the fifth partial derivative value is zero, the first weight matrix is determined.
- the object data set may be an image data set, and the objects in the object data set are images.
- the device provided by the embodiment of the present disclosure can execute the method provided by any embodiment of the present disclosure, and has corresponding functional modules and beneficial effects for executing the method.
- Each module in the above-mentioned apparatus for determining the object feature matrix can be fully or partially realized by software, hardware and a combination thereof.
- the above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, and can also be stored in the memory of the computer device in the form of software, so that the processor can invoke and execute the corresponding operations of the above-mentioned modules.
- FIG. 5 is a schematic structural diagram of an electronic device provided by an embodiment of the present disclosure.
- the electronic device includes a processor 501, a memory 502, an input device 503, and an output device 504; the number of processors 501 in the electronic device There may be one or more, and a processor 501 is taken as an example in FIG. connection as an example.
- the memory 502 as one or more non-volatile computer-readable storage media storing computer-readable instructions, can be used to store software programs, computer-executable programs, and modules, such as the program instructions/programs corresponding to the methods in the embodiments of the present disclosure. module.
- the processor 501 executes various functional applications and data processing of the electronic device by running software programs, instructions and modules stored in the memory 502 , that is, implements the methods provided by the embodiments of the present disclosure.
- the memory 502 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system and at least one application required by a function; the data storage area may store data created according to the use of the terminal, and the like.
- the memory 502 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other non-volatile solid-state storage devices.
- the memory 502 may further include a memory that is remotely located relative to the processor 501, and these remote memories may be connected to the computer device through a network. Examples of the aforementioned networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.
- the input device 503 can be configured to receive input digital or character information, and generate key signal input related to user settings and function control of the electronic device, and can include a keyboard, mouse, etc.
- the output device 504 can include a display device such as a display screen.
- an electronic device including a memory and one or more processors, the memory stores computer-readable instructions, and when the one or more processors execute the computer-readable instructions, the following steps are implemented: obtaining The heterogeneous feature correlation matrix and the corresponding correlation indicator matrix of various types of objects in the object dataset, as well as the multi-layer attribute heterogeneous network matrix of various types of objects; based on the heterogeneous feature correlation matrix and the corresponding correlation indicator matrix construction Heterogeneous feature correlation function; construct attribute heterogeneous network function based on multi-layer attribute heterogeneous network matrix; construct minimum objective function based on heterogeneous feature correlation function and attribute heterogeneous network function; calculate various types of The low-rank feature matrix of objects; the low-rank feature matrix is used to reconstruct the object feature matrix of various types of objects in the object dataset.
- the processor when the processor executes the computer-readable instructions, the following steps are also implemented: constructing the first constraint function based on the first weight matrix in the heterogeneous feature correlation function; constructing the second weight matrix based on the attribute heterogeneous network function
- the second constraint function based on the heterogeneous feature correlation function and the attribute heterogeneous network function, construct the minimization objective function, including: based on the heterogeneous feature correlation function, the attribute heterogeneous network function, the first constraint function and the second constraint function, construct Minimize the objective function.
- the processor when the processor executes the computer-readable instructions, the following steps are also implemented: constructing an unknown noise correlation constraint function; constructing The minimization objective function includes: constructing the minimization objective function based on heterogeneous feature correlation function, attribute heterogeneity network function, first constraint function, second constraint function and unknown noise correlation constraint function.
- the processor when the processor executes the computer-readable instructions, the following steps are further implemented: using the alternating direction multiplier method to solve the minimization objective function, and determine the low-rank characteristic matrix of various types of objects.
- the processor when the processor executes the computer-readable instructions, the following steps are also implemented: the first weight matrix, the second weight matrix, the low-rank feature matrix, and the different The qualitative correlation network matrix, and the basis matrix of each multi-layer attribute heterogeneous network matrix, until reaching the preset number of iterations or convergence.
- the processor when the processor executes the computer-readable instructions, the following steps are also implemented: setting the first weight matrix, the second weight matrix, the low-rank feature matrix and the heterogeneous correlation network matrix to fixed values, and the multi-layer attribute heterogeneity Find the first partial derivative of the base matrix of the prime network matrix, and determine the base matrix of the multi-layer attribute heterogeneous network matrix when the first partial derivative is zero; set the first weight matrix, the second weight matrix, the low-rank feature matrix and The base matrix of the heterogeneous network matrix with multi-layer attributes is a fixed value, and the second partial derivative is calculated for the heterogeneous correlation network matrix.
- the heterogeneous correlation network matrix is determined; the first weight matrix, the second The basic matrix of the two-weight matrix, the heterogeneous association network matrix and the multi-layer attribute heterogeneous network matrix is a fixed value, and the third partial derivative is calculated for the low-rank feature matrix.
- the third partial derivative value is zero, the low-rank feature matrix is determined.
- the processor executes the computer-readable instructions, the following steps are further implemented: the object data set is an image data set, and the objects in the object data set are images.
- the processor when the processor executes the computer-readable instructions, the following steps are also implemented: obtaining the original data of various types of objects in the object data set; processing the original data of various types of objects in the object data set to obtain the processed data ; Based on the processed data, the heterogeneous feature correlation matrix and the corresponding correlation indicator matrix of various types of objects in the object data set are obtained, as well as the multi-layer attribute heterogeneous network matrix of various types of objects.
- the processor executes the computer-readable instructions, the following steps are further implemented: adding the heterogeneous feature correlation function and the attribute heterogeneous network function to obtain the sum of the heterogeneous feature correlation function and the attribute heterogeneous network function ; According to the sum of the heterogeneous feature correlation function and the attribute heterogeneous network function, the minimum objective function is constructed.
- Embodiments of the present disclosure also provide a storage medium containing computer-executable instructions, and the computer-executable instructions are used to implement the method provided by the embodiments of the present disclosure when executed by one or more processors of a computer.
- the computer-executable instructions are not limited to the above method operations, and may also perform related operations in the method provided by any embodiment of the present disclosure.
- One or more non-volatile storage media storing computer-readable instructions, when the computer-readable instructions are executed by one or more processors, the one or more processors perform the following steps: Obtain various types of objects in the data set The heterogeneous feature correlation matrix of the object and the corresponding correlation relationship indicator matrix, and the multi-layer attribute heterogeneous network matrix of various types of objects; the heterogeneous feature correlation function is constructed based on the heterogeneous feature correlation matrix and the corresponding correlation relationship indicator matrix; Construct the attribute heterogeneous network function based on the multi-layer attribute heterogeneous network matrix; construct the minimum objective function based on the heterogeneous feature correlation function and attribute heterogeneous network function; calculate the low-rank feature matrix of various types of objects based on the minimum objective function ; using low-rank feature matrices to reconstruct object feature matrices for various types of objects in the object dataset.
- the following steps are further implemented: constructing the first constraint function based on the first weight matrix in the heterogeneous feature correlation function; constructing the first constraint function based on the second weight matrix in the attribute heterogeneous network function Construct the second constraint function; based on the heterogeneous feature correlation function and the attribute heterogeneous network function, construct the minimization objective function, including: based on the heterogeneous feature correlation function, the attribute heterogeneous network function, the first constraint function and the second constraint function, Construct the minimization objective function.
- the following steps are also implemented: constructing an unknown noise correlation constraint function; based on the heterogeneous feature correlation function, the attribute heterogeneous network function, the first constraint function and the second constraint function, Constructing the minimization objective function includes: constructing the minimization objective function based on heterogeneous feature correlation function, attribute heterogeneity network function, first constraint function, second constraint function and unknown noise correlation constraint function.
- the following steps are further implemented: using the alternating direction multiplier method to solve the minimization objective function, and determine the low-rank characteristic matrix of various types of objects.
- the following steps are also implemented: the first weight matrix, the second weight matrix, and the low-rank feature matrix in the minimization objective function are sequentially and alternately solved by using the alternating direction multiplier method, The heterogeneous association network matrix, and the base matrix of each multi-layer attribute heterogeneous network matrix, until reaching a preset number of iterations or convergence.
- the following steps are also implemented: setting the first weight matrix, the second weight matrix, the low-rank feature matrix and the heterogeneous correlation network matrix to fixed values, and for the multi-layer attribute Find the first partial derivative of the base matrix of the heterogeneous network matrix, and determine the base matrix of the multi-layer attribute heterogeneous network matrix when the first partial derivative is zero; set the first weight matrix, the second weight matrix, and the low-rank feature matrix and the base matrix of the multi-layer attribute heterogeneous network matrix are fixed values, and the second partial derivative is calculated for the heterogeneous correlation network matrix, and when the second partial derivative value is zero, the heterogeneous correlation network matrix is determined; the first weight matrix, The base matrix of the second weight matrix, heterogeneous association network matrix and multi-layer attribute heterogeneous network matrix is a fixed value, and the third partial derivative is calculated for the low-rank feature matrix.
- the low-rank feature is determined Matrix; set the base matrix of the first weight matrix, low-rank feature matrix, heterogeneous association network matrix, and multi-layer attribute heterogeneous network matrix to a fixed value, and calculate the fourth partial derivative for the second weight matrix, in the fourth partial derivative value When it is zero, determine the second weight matrix; set the base matrix of the second weight matrix, low-rank feature matrix, heterogeneous association network matrix and multi-layer attribute heterogeneous network matrix to a fixed value, and calculate the fifth bias of the first weight matrix Derivative, when the fifth partial derivative value is zero, determine the first weight matrix.
- the object data set is an image data set
- the objects in the object data set are images.
- the following steps are further implemented: obtaining the original data of various types of objects in the object data set; processing the original data of various types of objects in the object data set to obtain the processed Data; based on the processed data, the heterogeneous feature correlation matrix and the corresponding correlation indicator matrix of various types of objects in the object data set are obtained, as well as the multi-layer attribute heterogeneous network matrix of various types of objects.
- the following steps are further implemented: adding the heterogeneous feature correlation function to the attribute heterogeneous network function to obtain the relationship between the heterogeneous feature correlation function and the attribute heterogeneous network function and; according to the sum of the heterogeneous feature correlation function and the attribute heterogeneous network function, the minimum objective function is constructed.
- the object characteristic matrix determination method constructs the minimization objective function; calculates the low-rank characteristic matrix of various types of objects based on the minimization objective function, and then uses the low-rank characteristic matrix to reconstruct various types of objects in the object data set
- the object feature matrix of can reduce the information loss caused by homogeneous conversion in attribute heterogeneous networks, and has strong industrial applicability.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Primary Health Care (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- General Health & Medical Sciences (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
一种对象特征矩阵确定方法、装置、设备和存储介质。所述方法包括:获取对象数据集中各种类型对象的异质特征关联矩阵和对应的关联关系指示矩阵,以及各种类型对象的多层属性异质网络矩阵(S101);基于异质特征关联矩阵和对应的关联关系指示矩阵构建异质特征关联函数(S102);基于多层属性异质网络矩阵构建属性异质网络函数(S103);基于异质特征关联函数和属性异质网络函数,构建最小化目标函数(S104);基于最小化目标函数计算各种类型对象的低秩特征矩阵(S105);采用低秩特征矩阵重构对象数据集中各种类型对象的对象特征矩阵(S106)。上述方法可以降低属性异质网络因同质转换而造成信息损失。
Description
本公开要求于2021年10月22日提交中国专利局、申请号为202111234766.2、发明名称为“对象特征矩阵确定方法、装置、设备和存储介质”的中国专利申请的优先权,其全部内容通过引用结合在本公开中。
本公开涉及一种对象特征矩阵确定方法、装置、设备和存储介质。
对图像、生物分子、社交网站用户等对象进行有效分类依赖于建立有效的图像分类器。在采用判别模型进行对象分类的方法中,构建有效的对象分类器的前提是提取有效地表征样本对象特征的对象特征矩阵。
相关技术中提出基于矩阵分解确定对象的低秩特征矩阵,基于低秩特征矩阵构建样本对象的对象特征矩阵的方案。基于矩阵分解构建对象特征矩阵的方案可以保证异质数据源的内部结构。但是由于对象数据的不完整性以及模型假设和实验设计的局限性,现有方法仍然存在同质数据转换造成的信息损失及已知异质关联数据不完整的问题。
发明内容
(一)要解决的技术问题
现行技术中存在同质数据转换造成的信息损失及已知异质关联数据不完整的问题。
(二)技术方案
根据本公开公开的各种实施例,提供一种对象特征矩阵确定方法、装置、设备和存储介质。
第一方面,本公开实施例提供一种对象特征矩阵确定方法,包括:
获取对象数据集中各种类型对象的异质特征关联矩阵和对应的关联关系指示矩阵,以及各种类型对象的多层属性异质网络矩阵;
基于所述异质特征关联矩阵和对应的所述关联关系指示矩阵构建异质特征关联函数;
基于所述多层属性异质网络矩阵构建属性异质网络函数;
基于所述异质特征关联函数和所述属性异质网络函数,构建最小化目标函数;
基于所述最小化目标函数计算各种类型对象的低秩特征矩阵;
采用所述低秩特征矩阵重构所述对象数据集中各种类型对象的对象特征矩阵。
可选地,所述方法还包括:
基于所述异质特征关联函数中的第一权重矩阵构建第一约束函数;
基于所述属性异质网络函数中的第二权重矩阵构建第二约束函数;
基于所述异质特征关联函数和所述属性异质网络函数,构建最小化目标函数,包括:
基于所述异质特征关联函数、所述属性异质网络函数、所述第一约束函数和所述第二约束函数,构建所述最小化目标函数。
可选地,所述方法还包括:
还包括:构建未知噪声关联约束函数;
基于所述异质特征关联函数、所述属性异质网络函数、所述第一约束函数和所述第二约束函数,构建所述最小化目标函数,包括:
基于所述异质特征关联函数、所述属性异质网络函数、所述第一约束函数、所述第二约束函数和所述未知噪声关联约束函数,构建所述最小化目标函数。
可选地,所述基于所述最小化目标函数确定各种类型对象的低秩特征矩阵,包括:
采用交替方向乘子法对所述最小化目标函数进行求解,确定各种类型对象的低秩特征矩阵。
所述采用交替方向乘子法对所述最小化目标函数进行求解,包括:
采用所述交替方向乘子法依次交替求解所述最小化目标函数中的第一权重矩阵,第二权重矩阵,低秩特征矩阵,异质关联网络矩阵,以及各个多层属性异质网络矩阵的基矩阵,直至达到预设的迭代次数或者收敛。
可选地,采用所述交替方向乘子法依次交替求解所述最小化目标函数中的第一权重矩阵,第二权重矩阵,低秩特征矩阵,异质关联网络矩阵,以及多层属性异质网络矩阵的基矩阵,包括:
设置所述第一权重矩阵、所述第二权重矩阵、所述低秩特征矩阵和所述异质关联网络矩阵为固定值,对所述多层属性异质网络矩阵的基矩阵求第一偏导,在所述第一偏导值为零时,确定所述多层属性异质网络矩阵的基矩阵;
设置所述第一权重矩阵、所述第二权重矩阵、所述低秩特征矩阵和多层属性异质网络矩阵的基矩阵为固定值,对所述异质关联网络矩阵求第二偏导,在所述第二偏导值为零时,确定所述异质关联网络矩阵;
设置所述第一权重矩阵、所述第二权重矩阵、所述异质关联网络矩阵和多层属性异质网络矩阵的基矩阵为固定值,对所述低秩特征矩阵求第三偏导,在所述第三偏导值为零时,确定所述低秩特征矩阵;
设置所述第一权重矩阵、所述低秩特征矩阵、所述异质关联网络矩阵和所述多层属性异质网络矩阵的基矩阵为固定值,对所述第二权重矩阵求第四偏导,在所述第四偏导值为零时,确定所述第二权重矩阵;
设置所述第二权重矩阵、所述低秩特征矩阵、所述异质关联网络矩阵和所述多层属性异质网络矩阵的基矩阵为固定值,对所述第一权重矩阵求第五偏导,在所述第五偏导值为零时,确定所述第一权重矩阵。
可选地,所述对象数据集为图像数据集,所述对象数据集中的对象为图像。
第二方面,本公开实施例提供一种对象特征矩阵确定装置,包括:
矩阵获取单元,配置成获取对象数据集中各种类型对象的异质特征关联矩阵和对应的关联关系指示矩阵,以及各种类型对象的多层属性异质网络矩阵;
函数构建单元,配置成基于所述异质特征关联矩阵和对应的所述关联关系指示矩阵构建异质特征关联函数;基于所述多层属性异质网络矩阵构建属性异质网络函数;以及,基于所述异质特征关联函数和所述属性异质网络函数,构建最小化目标函数;
求解单元,配置成基于所述最小化目标函数计算各种类型对象的低秩特征矩阵;
对象特征矩阵确定单元,配置成采用所述低秩特征矩阵重构所述对象数据集中各种类型对象的对象特征矩阵。
可选地,所述函数构建单元还配置成基于所述异质特征关联函数中的第一权重矩阵构建第一约束函数;基于所述属性异质网络函数中的第二权重矩阵构建第二约束函数;基于所述异质特征关联函数、所述属性异质网络函数、所述第一约束函数和所述第二约束函数,构建所述最小化目标函数。
可选地,所述函数构建单元还配置成构建未知噪声关联约束函数;基于所述异质特征关联函数、所述属性异质网络函数、所述第一约束函数、所述第二约束函数和所述未知噪声关联约束函数,构建所述最小化目标函数。
可选地,所述求解单元具体配置成采用交替方向乘子法对所述最小化目标函数进行求解,确定各种类型对象的低秩特征矩阵。
可选地,所述求解单元具体配置成采用所述交替方向乘子法依次交替求解所述最小化目标函数中的第一权重矩阵,第二权重矩阵,低秩特征矩阵,异质关联网络矩阵,以及各个多层属性异质网络矩阵的基矩阵,直至达到预设的迭代次数或者收敛。
可选地,所述求解单元具体配置成设置所述第一权重矩阵、所述第二权重矩阵、所述低秩特征矩阵和所述异质关联网络矩阵为固定值,对所述多层属性异质网络矩阵的基矩阵求第一偏导,在所述第一偏导值为零时,确定所述多层属性异质网络矩阵的基矩阵;设置所述第一权重矩阵、所述第二权重矩阵、所述低秩特征矩阵和多层属性异质网络矩阵的基矩阵为固定值,对所述异质关联网络矩阵求第二偏导,在所述第二偏导值为零时,确定所述异质关联网络矩阵;设置所述第一权重矩阵、所述第二权重矩阵、所述异质关联网络矩阵和多层属性异质网络矩阵的基矩阵为固定值,对所述低秩特征矩阵求第三偏导,在所述第三偏导值为零时,确定所述低秩特征矩阵;设置所述第一权重矩阵、所述低秩特征矩阵、所述异质关联网络矩阵和所述多层属性异质网络矩阵的基矩阵为固定值,对所述第二权重矩阵求第四偏导,在所述第四偏导值为零时,确定所述第二权重矩阵;设置所述第二权重矩阵、所述低秩特征矩阵、所述异质关联网络矩阵和所述多层属性异质网络矩阵的基矩阵为固定值,对所述第一权重矩阵求第五偏导,在所述第五偏导值为零时,确定所述第一权重矩阵。
可选地,所述对象数据集为图像数据集,所述对象数据集中的对象为图像。
可选地,所述矩阵获取单元,具体配置成获取对象数据集中各种类型对象的原始数据;对所述对象数据集中各种类型对象的原始数据进行处理,得到处理后的数据;基于所述处理后的数据,获取所述对象数据集中各种类型对象的异质特征关联矩阵和对应的关联关系指示矩阵,以及各种类型对象的多层属性异质网络矩阵。
可选地,所述函数构建单元,具体配置成将所述异质特征关联函数与所述属性异质网络函数相加,得到所述异质特征关联函数与所述属性异质网络函数之和;根据所述异质特征关联函数与所述属性异质网络函数之和,构建最小化目标函数。
一种电子设备,包括:存储器和一个或多个处理器,所述存储器中存储有计算机可读指令;所述一个或多个处理器执行所述计算机可读指令时,使得所述一个或多个处理器执行本公开任意一个实施例中提供的对象特征矩阵确定方法的步骤。
一个或多个存储有计算机可读指令的非易失性计算机可读存储介质,计算机可读指令被一个或多个处理器执行时,使得一个或多个处理器执行本公开任意一个实施例中提供的对象特征矩阵确定方法的步骤。
此处的附图被并入说明书中并构成本说明书的一部分,示出了符合本公开的实施例,并与说明书一起用于解释本公开的原理。
为了更清楚地说明本公开实施例或现有技术中的技术方案,下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍,显而易见地,对于本领域普通技术人员而言,在不付出创造性劳动性的前提下,还可以根据这些附图获得其他的附图。
图1是本公开一个或多个实施例提供的对象特征矩阵确定方法流程图;
图2是本公开一个或多个实施例提供的一种对象特征矩阵确定的方法流程图;
图3是本公开一个或多个实施例提供的一种对象特征矩阵确定的方法流程图;
图4是本公开一个或多个实施例提供的一种对象确定装置的结构示意图;
图5是本公开一个或多个实施例提供的一种电子设备的结构示意图。
为了能够更清楚地理解本公开的上述目的、特征和优点,下面将对本公开的方案进行进一步描述。需要说明的是,在不冲突的情况下,本公开的实施例及实施例中的特征可以相互组合。
在下面的描述中阐述了很多具体细节以便于充分理解本公开,但本公开还可以采用其他不同于在此描述的方式来实施;显然,说明书中的实施例只是本公开的一部分实施例,而不是全部的实施例。
本公开实施例提供一种对象特征矩阵确定方法,用于解决现有基于矩阵分解确定对象特征的方法在融合(也就是矩阵分解)过程中因同质数据转换造成的信息损失,以及已知异质数据关联不完整的问题,继而更为准确地表征对象数据集中对象的特征。具体实施例中,本公开实施例中的对象可以是图像、生物分子、社交网站用户等类型的对象。
其中,本公开的对象特征矩阵确定方法由电子设备或者电子设备中的应用程序等来执行。电子设备可以是平板电脑、手机、笔记本电脑、服务器等设备,本公开对电子设备的具体类型不作任何限制。本公开对电子设备的操作系统的类型不做限定。例如,Android系统、Linux系统、Windows系统、iOS系统等。
图1是本公开实施例提供的对象特征矩阵确定方法流程图。如图1所示,本公开实施例提供的对象特征矩阵确定方法包括步骤S101-S106。
步骤S101:获取对象数据集中各种类型对象的异质特征关联矩阵和对应的关联关系指示矩阵,以及各种类型对象的多层属性异质网络矩阵。
本公开实施例中,电子设备可以对对象数据集中各种类型对象的原始数据进行处理,获取各种类型对象的异质特征关联矩阵、对应的关联关系指示矩阵和多层属性异质网络矩阵,也可以通过读取存储中预先存储的数据,获取前述矩阵。
本公开实施例采用
表示第i种类型对象的n
i个样本之间的异质特征关联矩阵,采用H
ij表示与R
ij对应的并且维度同R
ij相同的关联关系指示矩阵。其中关联关系指示矩阵H
ij用于将一致特征关联关系矩阵中的观察到的关联关系和未观察到的关联关系区分开。具体的,如果R
ij(s,t)>0,那么H
ij=1,否则H
ij=0。
步骤S102:基于异质特征关联矩阵和对应的关联关系指示矩阵构建异质特征关联函数。
异质特征关联函数中除了包括前述的异质特征关联矩阵和对应的关联管理指示矩阵外,还包括第一权重矩阵、各种类型对象对应的低秩特征矩阵和各种类型对象之间的异质关联网络矩阵。
在本公开的一个实施例中,异质特征关联函数采用公式1.1表示。
s.t.ω
r≥0,∑vec(ω
r)=1
其中:e表示哈达玛积;
表示第i种或第j种类型对象之间的异构关联网络指导的低秩特征矩阵,两个低秩标识矩阵分别表示了在压缩的k
i,k
j维空间以实数化的形式描述第i种类型对象的n
i个对象的属性信息和第j种类型对象的n
j个对象的属性信息;k
i和k
j分别表示第i种或者第j种低秩特征矩阵的维度;
表示了异质关联网络矩阵,其相比于R
ij规模小很多;
表示对
个异质关联网络所赋予的第一权重矩阵,对于
可近似为重构损失并被用于区别已观察到的关联关系和未观察到的关联关系,并使得已观察到的关联在重构后的R
ij中得以保持。
步骤S103:基于多层属性异质网络矩阵构建属性异质网络函数。
本公开实施例中,属性异质网络函数除了包括前述的多种属性异质网络矩阵外,还包括第二权重矩阵、各种类型对象对应的低秩特征矩阵和各种类型对象的属性异质网络分解后的基矩阵。
在本公开的一个实施例中,属性异质网络函数采用公式1.2表示。
本公开实施例中,采用属性异质网络函数可以直接将属性异构数据进行分解,避免同质转换造成的信息损失。
步骤S104:基于异质特征关联函数和属性异质网络函数,构建最小化目标函数。
本公司实施例中,基于异质特征关联函数和属性异质网络函数构建最小化目标函数,可以是将异质特征关联函数和属性异质网络函数相加。
在本公开的一个实施例中,在将异质特征关联函数和属性异质网络函数相加后,得到的最小化目标函数采用公式1.3表示
s.t.ω
r≥0,ω
h≥0,∑vec(ω
r)=1,∑vec(ω
h)=1 (1.3)
步骤S105:基于最小化目标函数计算各种类型对象的低秩特征矩阵。
本公开实施例中,基于最小化目标函数计算各种类型对象的低秩特征矩阵是对最小化目标函数中的第一权重矩阵、各种类型对象对应的低秩特征矩阵、各种类型对象之间的异质关联网络矩阵、第二权重矩阵、各种类型对象的属性异质网络分解后的基矩阵进行协同分解,并在协同分解过程中获取得到各种类型对象的低秩特征矩阵G。
本公开实施例中,最小化目标函数在G,S,U,ω
r,ω
h上是非凸的,因此可以借助被用于近似三因子矩阵分解的交替方向乘子法(Alternating Direction Method of Multipliers,ADMM)对其进行优化求解。具体的,可以将G,S,U,ω
r,ω
h中的四个参数设为常数,同时优化另一个,反复迭代,直至所有的参数都求解完成。具体采用交替方向乘子法计算G,S,U,ω
r,ω
h在下文中在做表述。
步骤S106:采用低秩特征矩阵重构对象数据集中各种类型对象的对象特征矩阵。
在确定各种类型对象的低秩特征矩阵后,可以采用低秩特征矩阵重构对象特征矩阵。在本公开一些实施例中,在确定各种类型对象的低秩特征矩阵后,可以直接将低秩特征矩阵作为对应类型对象的对象特征矩阵。
本公开实施提供的对象特征矩阵确定方法,通过异质特征关联矩阵和对应的关联关系指示矩阵构建异质特征关联函数,以实现利用关联关系指示矩阵对异质的异质关联和潜在的异质关联进行区分。通过采用各种类型对象对应的低秩特征矩阵和各种类型对象的属性异质网络分解后的基矩阵构建属性异质网络函数,并基于异质特征关联函数和属性异质网络函数构建最小化目标函数,利用最小化目标函数对异质特征关联函数和属性异质网络函数中的第一权重矩阵、低秩特征矩阵、异质关联网络矩阵、第二权重矩阵和各种类型对象的属性异质网络分解后的基矩阵协同分级,而获得各种类型对象的低秩特征矩阵,并基于各种类型对象的低秩特征矩阵确定各种类型对象的对象特征矩阵。
采用本公开实施例提供的方法,可以使得网络的拓扑结构、节点的属性信息融合在一起,从而弥补已知关联不足造成的冷启动问题,进而降低了属性异质网络因同质转换而造成信息损失。
图2是本公开实施例提供的一种对象特征矩阵确定的方法流程图。如图2所示,本公开实施例提供的方法包括步骤S201-S207。
步骤S201:获取对象数据集中各种类型对象的异质特征关联矩阵和对应的关联关系指示矩阵,以及各种类型对象的多层属性异质网络矩阵。
步骤S202:基于异质特征关联矩阵和对应的关联关系指示矩阵构建异质特征关联函数。
步骤S203:基于多层属性异质网络矩阵构建属性异质网络函数。
前述步骤S201-S203与前文实施例步骤S101-S103步骤相同,具体可以参见前文实施例,此处不再复述。
步骤S204:基于异质特征关联函数中的第一权重矩阵构建第一约束函数;以及,基于属性异质网络函数中的第二权重矩阵构建第二约束函数。
步骤S205:基于异质特征关联函数、属性异质网络函数、第一约束函数和第 二约束函数,构建最小化目标函数。
在对公式1.3所示的最小化目标函数进行求解时,当R
ij具有最小的近似损失
时,R
ij的权重将为
此时其他异质关联矩阵将全都被忽略。同样的,当X
it具有最少的近似损失
此时倾向于将
分配给X
it。也就是说,所有其他的同质矩阵的贡献都会被忽略。
同时大量的研究已经证实,不同的数据源彼此之间可以提供补充信息。因此,仅使用单个的异质关联矩阵和单个的同质关联矩阵可能无法给出可靠的预测。
为了弥补前述缺陷的权重分配,本公开实施例中首先基于异质特征关联函数中的第一权重矩阵构建第一约束函数,以及基于属性异质网络函数中的第二权重矩阵构建第二约束函数。
本公开实施例中,第一约束函数和第二约束函数均为基于l2范数的正则项。第一约束函数为
第二约束函数为
前述vec(ω
r)是将ω
r的行堆叠拼接后的向量,vec(ω
h)是将ω
h的行堆叠拼接后的向量。α>0,β>0被用来控制vec(ω
r)和vec(ω
h)的复杂度。同时,α,β还可以帮助选择性的集成不同的异质关联数据源和属性异质数据源。
在添加第一约束函数和第二约束函数的情况下,本公开实施例中得到的最小化目标函数采用公式1.4表示。
步骤S206:基于最小化目标函数计算各种类型对象的低秩特征矩阵。
步骤S207:采用所述低秩特征矩阵重构所述对象数据集中各种类型对象的对象特征矩阵。
前述步骤S206-S207与前文实施例步骤S105-S106步骤相同,具体可以参见前文实施例,此处不再复述。
本公开实施例提供的对象特征矩阵确定方法,基于异质特征关联函数中的第一权重矩阵构建第一约束函数,以及基于属性异质网络函数中的第二权重矩阵构建第二约束函数,通过第一约束函数和第二约束函数与异质特征关联函数、属性异质网络函数构建最小化目标函数,并基于最小化目标函数计算得到各个对象的特征,弥补了仅使用单个的异质关联矩阵和单个的同质关联矩阵可能无法给出可靠的预测的缺陷。
图3是本公开实施例提供的一种对象确定的方法流程图。如图3所示,本公开实施例提供的方法包括步骤S301-S308。
步骤S301:获取对象数据集中各种类型对象的异质特征关联矩阵和对应的关联关系指示矩阵,以及各种类型对象的多层属性异质网络矩阵。
步骤S302:基于异质特征关联矩阵和对应的关联关系指示矩阵构建异质特征关联函数。
步骤S303:基于多层属性异质网络矩阵构建属性异质网络函数。
步骤S304:基于异质特征关联函数中的第一权重矩阵构建第一约束函数;以及,基于属性异质网络函数中的第二权重矩阵构建第二约束函数。
前述步骤S301-S304与前文实施例步骤S201-S204步骤相同,具体可以参见前文实施例,此处不再复述。
步骤S305:构建未知噪声关联约束函数。
具体实施例中,关联关系指示矩阵H
ij将已观察到的关联关系和未观察到的关联关系进行区分,但是仅对已观察到的关联进行约束,从而很可能在某种程度上随机留下大量未观察的关联未加约束,进而导致目标关联矩阵中出现大量噪声。
为了避免前述问题,本公开实施例中,还会构建未知噪声关联约束函数,并基于未知噪声关联约束函数构建最小化目标函数。
应当注意的是,前述的步骤S304和S305可以顺序执行,也可以并行执行。
步骤S306:基于异质特征关联函数、属性异质网络函数、第一约束函数、第二约束函数和未知噪声关联约束函数,构建最小化目标函数。
将确定未知噪声关联约束函数后,本公开实施例中构建的最小化目标函数为
步骤S307:基于最小化目标函数计算各种类型对象的低秩特征矩阵。
步骤S308:采用低秩特征矩阵重构对象数据集中各种类型对象的对象特征矩阵。
前述步骤S307-S308与前文实施例步骤S206-S207步骤相同,具体可以参见前文实施例,此处不再复述。
前文提及,在本方案实施过程中,可以采用交替方向乘子法对最小化目标函数进行求解,确定各种类型对象的低秩特征矩阵。具体的,采用交替方向乘子法计算各种类型对象的低秩特征矩阵,为采用交替方向乘子法依次交替求解最小化目标函数中的第一权重矩阵,第二权重矩阵,低秩特征矩阵,异质关联网络矩阵,以及各个多层属性异质网络矩阵的基矩阵,直至达到预设的迭代次数或者收敛。
本公开实施例中,采用交替方向乘子法对最小化目标函数进行求解可以包括步骤S401-S405。以下以求解公式1.5为例,对本公开实施例采用的交替方向乘子算法进行分析。
随后可以基于公式1.6采用交替方向乘子算法执行步骤S401-S405。
步骤S401:设置第一权重矩阵、第二权重矩阵、低秩特征矩阵和异质关联网络矩阵为固定值,对多层属性异质网络矩阵的基矩阵求第一偏导,在第一偏导值为零时,确定多层属性异质网络矩阵的基矩阵。
具体的,假设G,S,ω
r,ω
h已知,可以优化U
it,因此对公式1.6求关于U
it的偏导数:
步骤S402:设置第一权重矩阵、第二权重矩阵、低秩特征矩阵和多层属性异质网络矩阵的基矩阵为固定值,对异质关联网络矩阵求第二偏导,在第二偏导值为零时,确定异质关联网络矩阵。
具体的,假设G,U,ω
r,ω
h已知,可以优化S
ij,因此对式1.6求关于S
ij的偏导数得到:
步骤S403:设置第一权重矩阵、第二权重矩阵、异质关联网络矩阵和多层属性异质网络矩阵的基矩阵为固定值,对低秩特征矩阵求第三偏导,在第三偏导值为零时,确定低秩特征矩阵。
具体的,假设S,U,ω
r,ω
h已知,可以对式1.6求关于G
i的偏导数:
其中e代表哈达玛积,公式1.10是一个定点方程且解必须满足收敛条件,因此可使得:
对于t=1,2,K,max
it
i:
步骤S404:设置第一权重矩阵、低秩特征矩阵、异质关联网络矩阵和多层属性异质网络矩阵的基矩阵为固定值,对第二权重矩阵求第四偏导,在第四偏导值为零时,确定第二权重矩阵。
具体的,在对ω
h求偏导时,公式1.6右侧的第1、3、5部分与ω
h无关,因此可以忽略。进而可得:
对于公式1.15的求解可以看做是关于vec(ω
h)的二次规划问题,可以基于选择性矩阵分解(Selective Non-Matrix Factorization,SNMF)的算法引入拉格朗日乘数来进行求解。
步骤S405:设置第二权重矩阵、低秩特征矩阵、异质关联网络矩阵和多层属性异质网络矩阵的基矩阵为固定值,对第一权重矩阵求第五偏导,在第五偏导值为零时,确定第一权重矩阵。
具体的,当对ω
r求偏导时,公式1.6右侧的第2、4、5部分与ω
r无关,因此可以忽略。进而可得:
公式1.17可以看做关于vec(ω
r)的二次规划问题,也可以基于选择性矩阵分解的算法引入拉格朗日乘数来进行求解。应当注意的是,前述的步骤S401-S405具体实施例中并没有顺序地限制。
在上述实施例的基础上,本公开实施例提供一种对象确定装置。图4是本公开实施例提供的一种对象确定装置的结构示意图。如图4所示,本公开实施例提供的对象确定装置包括矩阵获取单元401、函数构建单元402、求解单元403和对象特征矩阵确定单元404。
矩阵获取单元401配置成获取对象数据集中各种类型对象的异质特征关联矩阵和对应的关联关系指示矩阵,以及各种类型对象的多层属性异质网络矩阵;
函数构建单元402配置成基于异质特征关联矩阵和对应的关联关系指示矩阵构建异质特征关联函数;基于多层属性异质网络矩阵构建属性异质网络函数;以及,基于异质特征关联函数和属性异质网络函数,构建最小化目标函数;
求解单元403配置成基于最小化目标函数计算各种类型对象的低秩特征矩阵;
对象特征矩阵确定单元404配置成采用低秩特征矩阵重构对象数据集中各种类型对象的对象特征矩阵。
在本公开实施例提供的对象特征矩阵确定装置,通过异质特征关联矩阵和对应的关联关系指示矩阵构建异质特征关联函数,以实现利用关联关系指示矩阵对异质的异质关联和潜在的异质关联进行区分。通过采用各种类型对象对应的低秩特征矩阵和各种类型对象的属性异质网络分解后的基矩阵构建属性异质网络函数,并基于异质特征关联函数和属性异质网络函数构建最小化目标函数,利用最小化目标函数对异质特征关联函数和属性异质网络函数中的第一权重矩阵、低秩特征矩阵、异质关联网络矩阵、第二权重矩阵和各种类型对象的属性异质网络分解后的基矩阵协同分级,而获得各种类型对象的低秩特征矩阵,并基于各种类型对象的低秩特征矩阵确定各种类型对象的对象特征矩阵。
采用本公开实施例提供的装置,可以使得网络的拓扑结构、节点的属性信息融合在一起,从而弥补已知关联不足造成的冷启动问题,进而降低了属性异质网络因同质转换而造成信息损失。
在本公开一些实施例中,函数构建单元402还配置成基于异质特征关联函数中的第一权重矩阵构建第一约束函数,以及基于属性异质网络函数中的第二权重矩阵构建第二约束函数。对应的,函数构建单元402基于异质特征关联函数、属 性异质网络函数、第一约束函数和第二约束函数,构建最小化目标函数。
在本公开的一些实施例中,函数构建单元402还配置成构建未知噪声关联约束函数。对应的,函数构建单元402基于异质特征关联函数、属性异质网络函数、第一约束函数、第二约束函数和未知噪声关联约束函数,构建最小化目标函数。
在本公开一些实施例中,求解单元403采用交替方向乘子法对最小化目标函数进行求解,确定各种类型对象的低秩特征矩阵。
具体的,在本公开的一些实施例中,求解单元403采用交替方向乘子法依次交替求解最小化目标函数中的第一权重矩阵,第二权重矩阵,低秩特征矩阵,异质关联网络矩阵,以及各个多层属性异质网络矩阵的基矩阵,直至达到预设的迭代次数或者收敛。
实际实施中,求解单元403通过设置第一权重矩阵、第二权重矩阵、低秩特征矩阵和异质关联网络矩阵为固定值,对多层属性异质网络矩阵的基矩阵求第一偏导,在第一偏导值为零时,确定多层属性异质网络矩阵的基矩阵;
实际实施中,求解单元403通过设置第一权重矩阵、第二权重矩阵、低秩特征矩阵和多层属性异质网络矩阵的基矩阵为固定值,对异质关联网络矩阵求第二偏导,在第二偏导值为零时,确定异质关联网络矩阵;
实际实施中,求解单元403通过设置第一权重矩阵、第二权重矩阵、异质关联网络矩阵和多层属性异质网络矩阵的基矩阵为固定值,对低秩特征矩阵求第三偏导,在第三偏导值为零时,确定低秩特征矩阵;
实际实施中,求解单元403通过设置第一权重矩阵、低秩特征矩阵、异质关联网络矩阵和多层属性异质网络矩阵的基矩阵为固定值,对第二权重矩阵求第四偏导,在第四偏导值为零时,确定第二权重矩阵;
实际实施中,求解单元403通过设置第二权重矩阵、低秩特征矩阵、异质关联网络矩阵和多层属性异质网络矩阵的基矩阵为固定值,对第一权重矩阵求第五偏导,在第五偏导值为零时,确定第一权重矩阵。
本公开实施例中,对象数据集可以为图像数据集,对象数据集中的对象为图像。
本公开实施例所提供的装置可执行本公开任意实施例所提供的方法,具备执行方法相应的功能模块和有益效果。关于对象特征矩阵确定装置的具体限定可以参见上文中对于对象特征矩阵确定方法的限定,在此不再赘述。上述对象特征矩阵确定装置中的各个模块可全部或部分通过软件、硬件及其组合来实现。上述各模块可以硬件形式内嵌于或独立于计算机设备中的处理器中,也可以以软件形式存储于计算机设备中的存储器中,以便于处理器调用执行以上各个模块对应的操作。
图5是本公开实施例提供的一种电子设备的结构示意图,如图5所示,该电子设备包括处理器501、存储器502、输入装置503和输出装置504;电子设备中处理器501的数量可以是一个或多个,图5中以一个处理器501为例;电子设备中的处理器501、存储器502、输入装置503和输出装置504可以通过总线或其他方式连接,图5中以通过总线连接为例。
存储器502作为一个或多个存储有计算机可读指令的非易失性计算机可读存储介质,可用于存储软件程序、计算机可执行程序以及模块,如本公开实施例中 的方法对应的程序指令/模块。处理器501通过运行存储在存储器502中的软件程序、指令以及模块,从而执行电子设备的各种功能应用以及数据处理,即实现本公开实施例所提供的方法。
存储器502可主要包括存储程序区和存储数据区,其中,存储程序区可存储操作系统、至少一个功能所需的应用程序;存储数据区可存储根据终端的使用所创建的数据等。此外,存储器502可以包括高速随机存取存储器,还可以包括非易失性存储器,例如至少一个磁盘存储器件、闪存器件、或其他非易失性固态存储器件。在一些实例中,存储器502可进一步包括相对于处理器501远程设置的存储器,这些远程存储器可以通过网络连接至计算机设备。上述网络的实例包括但不限于互联网、企业内部网、局域网、移动通信网及其组合。
输入装置503可配置成接收输入的数字或字符信息,以及产生与电子设备的用户设置以及功能控制有关的键信号输入,可以包括键盘、鼠标等,输出装置504可包括显示屏等显示设备。
在一个实施例中,提供了一种电子设备,包括存储器和一个或多个处理器,该存储器存储有计算机可读指令,一个或多个处理器执行计算机可读指令时,实现以下步骤:获取对象数据集中各种类型对象的异质特征关联矩阵和对应的关联关系指示矩阵,以及各种类型对象的多层属性异质网络矩阵;基于该异质特征关联矩阵和对应的关联关系指示矩阵构建异质特征关联函数;基于多层属性异质网络矩阵构建属性异质网络函数;基于异质特征关联函数和属性异质网络函数,构建最小化目标函数;基于该最小化目标函数计算各种类型对象的低秩特征矩阵;采用低秩特征矩阵重构对象数据集中各种类型对象的对象特征矩阵。
在一个实施例中,处理器执行计算机可读指令时还实现以下步骤:基于异质特征关联函数中的第一权重矩阵构建第一约束函数;基于属性异质网络函数中的第二权重矩阵构建第二约束函数;基于异质特征关联函数和属性异质网络函数,构建最小化目标函数,包括:基于异质特征关联函数、属性异质网络函数、第一约束函数和第二约束函数,构建最小化目标函数。
在一个实施例中,处理器执行计算机可读指令时还实现以下步骤:构建未知噪声关联约束函数;基于异质特征关联函数、属性异质网络函数、第一约束函数和第二约束函数,构建最小化目标函数,包括:基于异质特征关联函数、属性异质网络函数、第一约束函数、第二约束函数和未知噪声关联约束函数,构建最小化目标函数。
在一个实施例中,处理器执行计算机可读指令时还实现以下步骤:采用交替方向乘子法对最小化目标函数进行求解,确定各种类型对象的低秩特征矩阵。
在一个实施例中,处理器执行计算机可读指令时还实现以下步骤:采用交替方向乘子法依次交替求解最小化目标函数中的第一权重矩阵,第二权重矩阵,低秩特征矩阵,异质关联网络矩阵,以及各个多层属性异质网络矩阵的基矩阵,直至达到预设的迭代次数或者收敛。
在一个实施例中,处理器执行计算机可读指令时还实现以下步骤:设置第一 权重矩阵、第二权重矩阵、低秩特征矩阵和异质关联网络矩阵为固定值,对该多层属性异质网络矩阵的基矩阵求第一偏导,在第一偏导值为零时,确定多层属性异质网络矩阵的基矩阵;设置第一权重矩阵、第二权重矩阵、低秩特征矩阵和多层属性异质网络矩阵的基矩阵为固定值,对异质关联网络矩阵求第二偏导,在第二偏导值为零时,确定异质关联网络矩阵;设置第一权重矩阵、第二权重矩阵、异质关联网络矩阵和多层属性异质网络矩阵的基矩阵为固定值,对低秩特征矩阵求第三偏导,在第三偏导值为零时,确定低秩特征矩阵;设置第一权重矩阵、低秩特征矩阵、异质关联网络矩阵和多层属性异质网络矩阵的基矩阵为固定值,对第二权重矩阵求第四偏导,在第四偏导值为零时,确定第二权重矩阵;设置第二权重矩阵、低秩特征矩阵、异质关联网络矩阵和多层属性异质网络矩阵的基矩阵为固定值,对第一权重矩阵求第五偏导,在第五偏导值为零时,确定第一权重矩阵。
在一个实施例中,处理器执行计算机可读指令时还实现以下步骤:对象数据集为图像数据集,对象数据集中的对象为图像。
在一个实施例中,处理器执行计算机可读指令时还实现以下步骤:获取对象数据集中各种类型对象的原始数据;对对象数据集中各种类型对象的原始数据进行处理,得到处理后的数据;基于处理后的数据,获取对象数据集中各种类型对象的异质特征关联矩阵和对应的关联关系指示矩阵,以及各种类型对象的多层属性异质网络矩阵。
在一个实施例中,处理器执行计算机可读指令时还实现以下步骤:将异质特征关联函数与属性异质网络函数相加,得到异质特征关联函数与所述属性异质网络函数之和;根据异质特征关联函数与属性异质网络函数之和,构建最小化目标函数。
本公开实施例还提供了一种包含计算机可执行指令的存储介质,计算机可执行指令在由计算机一个或多个处理器执行时用于实现本公开实施例所提供的方法。
当然,本公开实施例所提供的一种包含计算机可执行指令的存储介质,其计算机可执行指令不限于如上的方法操作,还可以执行本公开任意实施例所提供的方法中的相关操作。
一个或多个存储有计算机可读指令的非易失性存储介质,计算机可读指令被一个或多个处理器执行时,使得一个或多个处理器执行以下步骤:获取对象数据集中各种类型对象的异质特征关联矩阵和对应的关联关系指示矩阵,以及各种类型对象的多层属性异质网络矩阵;基于该异质特征关联矩阵和对应的关联关系指示矩阵构建异质特征关联函数;基于多层属性异质网络矩阵构建属性异质网络函数;基于异质特征关联函数和属性异质网络函数,构建最小化目标函数;基于该最小化目标函数计算各种类型对象的低秩特征矩阵;采用低秩特征矩阵重构对象数据集中各种类型对象的对象特征矩阵。
在一个实施例中,计算机可读指令被处理器执行时还实现以下步骤:基于异质特征关联函数中的第一权重矩阵构建第一约束函数;基于属性异质网络函数中的第二权重矩阵构建第二约束函数;基于异质特征关联函数和属性异质网络函数,构建最小化目标函数,包括:基于异质特征关联函数、属性异质网络函数、第一约束函数和第二约束函数,构建最小化目标函数。
在一个实施例中,计算机可读指令被处理器执行时还实现以下步骤:构建未知噪声关联约束函数;基于异质特征关联函数、属性异质网络函数、第一约束函数和第二约束函数,构建最小化目标函数,包括:基于异质特征关联函数、属性异质网络函数、第一约束函数、第二约束函数和未知噪声关联约束函数,构建最小化目标函数。
在一个实施例中,计算机可读指令被处理器执行时还实现以下步骤:采用交替方向乘子法对最小化目标函数进行求解,确定各种类型对象的低秩特征矩阵。
在一个实施例中,计算机可读指令被处理器执行时还实现以下步骤:采用交替方向乘子法依次交替求解最小化目标函数中的第一权重矩阵,第二权重矩阵,低秩特征矩阵,异质关联网络矩阵,以及各个多层属性异质网络矩阵的基矩阵,直至达到预设的迭代次数或者收敛。
在一个实施例中,计算机可读指令被处理器执行时还实现以下步骤:设置第一权重矩阵、第二权重矩阵、低秩特征矩阵和异质关联网络矩阵为固定值,对该多层属性异质网络矩阵的基矩阵求第一偏导,在第一偏导值为零时,确定多层属性异质网络矩阵的基矩阵;设置第一权重矩阵、第二权重矩阵、低秩特征矩阵和多层属性异质网络矩阵的基矩阵为固定值,对异质关联网络矩阵求第二偏导,在第二偏导值为零时,确定异质关联网络矩阵;设置第一权重矩阵、第二权重矩阵、异质关联网络矩阵和多层属性异质网络矩阵的基矩阵为固定值,对低秩特征矩阵求第三偏导,在第三偏导值为零时,确定低秩特征矩阵;设置第一权重矩阵、低秩特征矩阵、异质关联网络矩阵和多层属性异质网络矩阵的基矩阵为固定值,对第二权重矩阵求第四偏导,在第四偏导值为零时,确定第二权重矩阵;设置第二权重矩阵、低秩特征矩阵、异质关联网络矩阵和多层属性异质网络矩阵的基矩阵为固定值,对第一权重矩阵求第五偏导,在第五偏导值为零时,确定第一权重矩阵。
在一个实施例中,计算机可读指令被处理器执行时还实现以下步骤:对象数据集为图像数据集,对象数据集中的对象为图像。
在一个实施例中,计算机可读指令被处理器执行时还实现以下步骤:获取对象数据集中各种类型对象的原始数据;对对象数据集中各种类型对象的原始数据进行处理,得到处理后的数据;基于处理后的数据,获取对象数据集中各种类型对象的异质特征关联矩阵和对应的关联关系指示矩阵,以及各种类型对象的多层属性异质网络矩阵。
在一个实施例中,计算机可读指令被处理器执行时还实现以下步骤:将异质特征关联函数与属性异质网络函数相加,得到异质特征关联函数与所述属性异质网络函数之和;根据异质特征关联函数与属性异质网络函数之和,构建最小化目标函数。
需要说明的是,在本文中,诸如“第一”和“第二”等之类的关系术语仅仅用来将一个实体或者操作与另一个实体或操作区分开来,而不一定要求或者暗示这些实体或操作之间存在任何这种实际的关系或者顺序。而且,术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含,从而使得包括一系列要素的过程、方法、物品或者设备不仅包括那些要素,而且还包括没有明确列出的其他要素,或者是还包括为这种过程、方法、物品或者设备所固有的要素。在没有更多限制的情况下,由语句“包括一个……”限定的要素,并不排除在包括要素的过程、方法、物品或者设备中还存在另外的相同要素。
以上仅是本公开的具体实施方式,使本领域技术人员能够理解或实现本公开。对这些实施例的多种修改对本领域的技术人员来说将是显而易见的,本文中所定义的一般原理可以在不脱离本公开的精神或范围的情况下,在其它实施例中实现。因此,本公开将不会被限制于本文的这些实施例,而是要符合与本文所公开的原理和新颖特点相一致的最宽的范围。
本公开实施例提供的对象特征矩阵确定方法,通过构建最小化目标函数;基于最小化目标函数计算各种类型对象的低秩特征矩阵,进而采用低秩特征矩阵重构对象数据集中各种类型对象的对象特征矩阵,可以降低属性异质网络因同质转换而造成信息损失,具有很强的工业实用性。
Claims (20)
- 一种对象特征矩阵确定方法,其特征在于,包括:获取对象数据集中各种类型对象的异质特征关联矩阵和对应的关联关系指示矩阵,以及各种类型对象的多层属性异质网络矩阵;基于所述异质特征关联矩阵和对应的所述关联关系指示矩阵构建异质特征关联函数;基于所述多层属性异质网络矩阵构建属性异质网络函数;基于所述异质特征关联函数和所述属性异质网络函数,构建最小化目标函数;基于所述最小化目标函数计算各种类型对象的低秩特征矩阵;采用所述低秩特征矩阵重构所述对象数据集中各种类型对象的对象特征矩阵。
- 根据权利要求1所述的方法,其中,还包括:基于所述异质特征关联函数中的第一权重矩阵构建第一约束函数;基于所述属性异质网络函数中的第二权重矩阵构建第二约束函数;基于所述异质特征关联函数和所述属性异质网络函数,构建最小化目标函数,包括:基于所述异质特征关联函数、所述属性异质网络函数、所述第一约束函数和所述第二约束函数,构建所述最小化目标函数。
- 根据权利要求2所述的方法,其中,还包括:构建未知噪声关联约束函数;基于所述异质特征关联函数、所述属性异质网络函数、所述第一约束函数和所述第二约束函数,构建所述最小化目标函数,包括:基于所述异质特征关联函数、所述属性异质网络函数、所述第一约束函数、所述第二约束函数和所述未知噪声关联约束函数,构建所述最小化目标函数。
- 根据权利要求1-3任一项所述的方法,其中,所述基于所述最小化目标函数确定各种类型对象的低秩特征矩阵,包括:采用交替方向乘子法对所述最小化目标函数进行求解,确定各种类型对象的低秩特征矩阵。
- 根据权利要求4所述的方法,其中,所述采用交替方向乘子法对所述最小化目标函数进行求解,包括:采用所述交替方向乘子法依次交替求解所述最小化目标函数中的第一权重矩阵,第二权重矩阵,低秩特征矩阵,异质关联网络矩阵,以及各个多层属性异质网络矩阵的基矩阵,直至达到预设的迭代次数或者收敛。
- 根据权利要求5所述的方法,其中,采用所述交替方向乘子法依次交替求解所述最小化目标函数中的第一权重矩阵,第二权重矩阵,低秩特征矩阵,异质关联网络矩阵,以及多层属性异质网络矩阵的基矩阵,包括:设置所述第一权重矩阵、所述第二权重矩阵、所述低秩特征矩阵和所述异质 关联网络矩阵为固定值,对所述多层属性异质网络矩阵的基矩阵求第一偏导,在所述第一偏导值为零时,确定所述多层属性异质网络矩阵的基矩阵;设置所述第一权重矩阵、所述第二权重矩阵、所述低秩特征矩阵和多层属性异质网络矩阵的基矩阵为固定值,对所述异质关联网络矩阵求第二偏导,在所述第二偏导值为零时,确定所述异质关联网络矩阵;设置所述第一权重矩阵、所述第二权重矩阵、所述异质关联网络矩阵和多层属性异质网络矩阵的基矩阵为固定值,对所述低秩特征矩阵求第三偏导,在所述第三偏导值为零时,确定所述低秩特征矩阵;设置所述第一权重矩阵、所述低秩特征矩阵、所述异质关联网络矩阵和所述多层属性异质网络矩阵的基矩阵为固定值,对所述第二权重矩阵求第四偏导,在所述第四偏导值为零时,确定所述第二权重矩阵;设置所述第二权重矩阵、所述低秩特征矩阵、所述异质关联网络矩阵和所述多层属性异质网络矩阵的基矩阵为固定值,对所述第一权重矩阵求第五偏导,在所述第五偏导值为零时,确定所述第一权重矩阵。
- 根据权利要求1-3任一项所述的方法,其中,所述对象数据集为图像数据集,所述对象数据集中的对象为图像。
- 根据权利要求1所述的方法,其中,所述获取对象数据集中各种类型对象的异质特征关联矩阵和对应的关联关系指示矩阵,以及各种类型对象的多层属性异质网络矩阵,包括:获取对象数据集中各种类型对象的原始数据;对所述对象数据集中各种类型对象的原始数据进行处理,得到处理后的数据;基于所述处理后的数据,获取所述对象数据集中各种类型对象的异质特征关联矩阵和对应的关联关系指示矩阵,以及各种类型对象的多层属性异质网络矩阵。
- 根据权利要求1所述的方法,其中,所述基于所述异质特征关联函数和所述属性异质网络函数,构建最小化目标函数,包括:将所述异质特征关联函数与所述属性异质网络函数相加,得到所述异质特征关联函数与所述属性异质网络函数之和;根据所述异质特征关联函数与所述属性异质网络函数之和,构建最小化目标函数。
- 一种对象特征矩阵确定装置,其特征在于,包括:矩阵获取单元,配置成获取对象数据集中各种类型对象的异质特征关联矩阵和对应的关联关系指示矩阵,以及各种类型对象的多层属性异质网络矩阵;函数构建单元,配置成基于所述异质特征关联矩阵和对应的所述关联关系指示矩阵构建异质特征关联函数;基于所述多层属性异质网络矩阵构建属性异质网络函数;以及,基于所述异质特征关联函数和所述属性异质网络函数,构建最小化目标函数;求解单元,配置成基于所述最小化目标函数计算各种类型对象的低秩特征矩阵;对象特征矩阵确定单元,配置成采用所述低秩特征矩阵重构所述对象数据集中各种类型对象的对象特征矩阵。
- 根据权利要求10所述的装置,其中,所述函数构建单元还配置成基于所述异质特征关联函数中的第一权重矩阵构建第一约束函数;基于所述属性异质网络函数中的第二权重矩阵构建第二约束函数;基于所述异质特征关联函数、所述属性异质网络函数、所述第一约束函数和所述第二约束函数,构建所述最小化目标函数。
- 根据权利要求11所述的装置,其中,所述函数构建单元还配置成构建未知噪声关联约束函数;基于所述异质特征关联函数、所述属性异质网络函数、所述第一约束函数、所述第二约束函数和所述未知噪声关联约束函数,构建所述最小化目标函数。
- 根据权利要求10-12任一项所述的装置,其中,所述求解单元具体配置成采用交替方向乘子法对所述最小化目标函数进行求解,确定各种类型对象的低秩特征矩阵。
- 根据权利要求13所述的装置,其中,所述求解单元具体配置成采用所述交替方向乘子法依次交替求解所述最小化目标函数中的第一权重矩阵,第二权重矩阵,低秩特征矩阵,异质关联网络矩阵,以及各个多层属性异质网络矩阵的基矩阵,直至达到预设的迭代次数或者收敛。
- 根据权利要求14所述的装置,其中,所述求解单元具体配置成设置所述第一权重矩阵、所述第二权重矩阵、所述低秩特征矩阵和所述异质关联网络矩阵为固定值,对所述多层属性异质网络矩阵的基矩阵求第一偏导,在所述第一偏导值为零时,确定所述多层属性异质网络矩阵的基矩阵;设置所述第一权重矩阵、所述第二权重矩阵、所述低秩特征矩阵和多层属性异质网络矩阵的基矩阵为固定值,对所述异质关联网络矩阵求第二偏导,在所述第二偏导值为零时,确定所述异质关联网络矩阵;设置所述第一权重矩阵、所述第二权重矩阵、所述异质关联网络矩阵和多层属性异质网络矩阵的基矩阵为固定值,对所述低秩特征矩阵求第三偏导,在所述第三偏导值为零时,确定所述低秩特征矩阵;设置所述第一权重矩阵、所述低秩特征矩阵、所述异质关联网络矩阵和所述多层属性异质网络矩阵的基矩阵为固定值,对所述第二权重矩阵求第四偏导,在所述第四偏导值为零时,确定所述第二权重矩阵;设置所述第二权重矩阵、所述低秩特征矩阵、所述异质关联网络矩阵和所述多层属性异质网络矩阵的基矩阵为固定值,对所述第一权重矩阵求第五偏导,在所述第五偏导值为零时,确定所述第一权重矩阵。
- 根据权利要求10-12任一项所述的装置,其中,所述对象数据集为图像数据集,所述对象数据集中的对象为图像。
- 根据权利要求10所述的装置,其中,所述矩阵获取单元,具体配置成获取对象数据集中各种类型对象的原始数据;对所述对象数据集中各种类型对象的原始数据进行处理,得到处理后的数据;基于所述处理后的数据,获取所述对象数据集中各种类型对象的异质特征关联矩阵和对应的关联关系指示矩阵,以及各种类型对象的多层属性异质网络矩阵。
- 根据权利要求10所述的装置,其中,所述函数构建单元,具体配置成将所述异质特征关联函数与所述属性异质网络函数相加,得到所述异质特征关联函数与所述属性异质网络函数之和;根据所述异质特征关联函数与所述属性异质网络函数之和,构建最小化目标函数。
- 一种电子设备,包括:存储器和一个或多个处理器,所述存储器中存储有计算机可读指令;所述一个或多个处理器执行所述计算机可读指令时,使得所述一个或多个处理器执行权利要求1-9任一项所述的对象特征矩阵确定方法的步骤。
- 一个或多个存储有计算机可读指令的非易失性计算机可读存储介质,所述计算机可读指令被一个或多个处理器执行时,使得所述一个或多个处理器执行权利要求1-9任一项所述的对象特征矩阵确定方法的步骤。
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111234766.2 | 2021-10-22 | ||
CN202111234766.2A CN114036330A (zh) | 2021-10-22 | 2021-10-22 | 对象特征矩阵确定方法、装置、设备和存储介质 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2023065525A1 true WO2023065525A1 (zh) | 2023-04-27 |
Family
ID=80135208
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2021/140980 WO2023065525A1 (zh) | 2021-10-22 | 2021-12-23 | 对象特征矩阵确定方法、装置、设备和存储介质 |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN114036330A (zh) |
WO (1) | WO2023065525A1 (zh) |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20130191425A1 (en) * | 2012-01-20 | 2013-07-25 | Fatih Porikli | Method for Recovering Low-Rank Matrices and Subspaces from Data in High-Dimensional Matrices |
US20190050372A1 (en) * | 2017-08-14 | 2019-02-14 | City University Of Hong Kong | Systems and methods for robust low-rank matrix approximation |
CN110222213A (zh) * | 2019-05-28 | 2019-09-10 | 天津大学 | 一种基于异构张量分解的图像分类方法 |
CN111310813A (zh) * | 2020-02-07 | 2020-06-19 | 广东工业大学 | 一种潜在低秩表示的子空间聚类方法及装置 |
-
2021
- 2021-10-22 CN CN202111234766.2A patent/CN114036330A/zh active Pending
- 2021-12-23 WO PCT/CN2021/140980 patent/WO2023065525A1/zh unknown
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20130191425A1 (en) * | 2012-01-20 | 2013-07-25 | Fatih Porikli | Method for Recovering Low-Rank Matrices and Subspaces from Data in High-Dimensional Matrices |
US20190050372A1 (en) * | 2017-08-14 | 2019-02-14 | City University Of Hong Kong | Systems and methods for robust low-rank matrix approximation |
CN110222213A (zh) * | 2019-05-28 | 2019-09-10 | 天津大学 | 一种基于异构张量分解的图像分类方法 |
CN111310813A (zh) * | 2020-02-07 | 2020-06-19 | 广东工业大学 | 一种潜在低秩表示的子空间聚类方法及装置 |
Non-Patent Citations (1)
Title |
---|
YU GUOXIAN, WANG YUEHUI, WANG JUN, DOMENICONI CARLOTTA, GUO MAOZU, ZHANG XIANGLIANG: "Attributed heterogeneous network fusion via collaborative matrix tri-factorization", INFORMATION FUSION, ELSEVIER, US, vol. 63, 1 November 2020 (2020-11-01), US , pages 153 - 165, XP093057696, ISSN: 1566-2535, DOI: 10.1016/j.inffus.2020.06.012 * |
Also Published As
Publication number | Publication date |
---|---|
CN114036330A (zh) | 2022-02-11 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Crane et al. | Edge exchangeable models for interaction networks | |
US11860675B2 (en) | Latent network summarization | |
US7996762B2 (en) | Correlative multi-label image annotation | |
CN112714032B (zh) | 无线网络协议知识图谱构建分析方法、系统、设备及介质 | |
CN111400504B (zh) | 企业关键人的识别方法和装置 | |
BR112019008055A2 (pt) | método implementado por computador, meio não transitório, legível por computador e sistema implementado por computador | |
Kazienko et al. | Multidimensional social network: model and analysis | |
US20170308620A1 (en) | Making graph pattern queries bounded in big graphs | |
US10789240B2 (en) | Duplicative data detection | |
Liu et al. | Research of animals image semantic segmentation based on deep learning | |
US20240134892A1 (en) | Information generation method, apparatus, electronic device, and computer -readable medium | |
WO2024179409A1 (zh) | 三维工业异常检测方法、装置、存储介质及电子设备 | |
WO2024179409A9 (zh) | 三维工业异常检测方法、装置、存储介质及电子设备 | |
CN114637857A (zh) | 一种基于去噪的知识感知图卷积网络推荐方法 | |
WO2023065525A1 (zh) | 对象特征矩阵确定方法、装置、设备和存储介质 | |
WO2024139666A1 (zh) | 双目标域推荐模型的训练方法及装置 | |
CN110730191A (zh) | 基于数据、信息和知识对象的意图导向的osi七层网络协议模型 | |
CN107729821B (zh) | 一种基于一维序列学习的视频概括方法 | |
Zhang et al. | On streaming disaster damage assessment in social sensing: A crowd-driven dynamic neural architecture searching approach | |
CN113538413A (zh) | 图像检测方法及装置、电子设备和存储介质 | |
CN117056597A (zh) | 基于噪声增强的对比学习图推荐方法 | |
Arnold et al. | Likelihood-based approach to discriminate mixtures of network models that vary in time | |
CN116665228A (zh) | 图像处理方法及装置 | |
CN111784402A (zh) | 基于多通路的下单率预测方法、设备及可读存储介质 | |
CN114329099B (zh) | 重叠社区识别方法、装置、设备、存储介质及程序产品 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 21961263 Country of ref document: EP Kind code of ref document: A1 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |