Multi-view clustering method, device and equipment based on consistency and inconsistency
Through self-expression learning and HSIC analysis combined with Laplace rank constraints, the multi-view clustering method is solved, and the problem of inconsistency analysis and similar graph relationship neglect in multi-view clustering is improved, and the accuracy and reliability of clustering effect are improved. .
Patent Information
- Application Number
- CN202411174459.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-26
- Publication Date
- 2024-11-08
AI Technical Summary
The existing multi-view clustering method ignores the diversity and independent relationship of inconsistencies, resulting in poor clustering effects and separation of similar graph structures from clustering results.
The initial cluster similarity graph matrix is constructed using a multi-view subspace learning algorithm based on self-expression, and the consistency and inconsistency parts are fused through matrix decomposition and HSIC dependency analysis, and cluster analysis is performed in combination with Laplace rank constraints.
Effectively remove data noise, satisfy the neglected inconsistency analysis in multi-view clustering, ensure the accuracy and reliability of clustering results, and correlation relationship analysis and clustering results.
Smart Images

Figure CN118916723A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of data processing, and in particular to a multi-view clustering method, device and equipment based on consistency and inconsistency. Background Art
[0002] Technologies such as the Internet, big data, and artificial intelligence are developing rapidly in today's society, and with it comes an explosive growth in data. Data in various fields are characterized by massive, multi-source heterogeneity, and dynamic changes. In order to divide these high-dimensional and complex multi-view data into different clusters, multi-view clustering came into being and has become one of the research hotspots in the academic field. Multi-view clustering aims to integrate multi-view data with complex structures through certain methods, and then divide them reasonably and accurately. Therefore, how to efficiently mine the structural information of multi-view data to obtain a unified data representation, and then obtain appropriate data grouping.
[0003] Graph matrix learning is a good choice for exploring the complex structural relationships between multi-view data. Therefore, multi-view clustering based on graph matrix learning has received widespread attention. It aims to build a similarity matrix of data, then fuse the similarity matrix of each view, and finally use a typical spectral clustering algorithm or other methods to perform the final clustering on the fused graph matrix, showing strong superiority in improving clustering performance.
[0004] The purpose of multi-view learning is to effectively mine the associated information of each view and follow the principles of complementarity and consistency. For each view, there must be consistent parts and inconsistent parts. The existing multi-view clustering analysis of consistency and inconsistency ignores the diversity and mutual independence of inconsistencies. Moreover, the correlation between multiple different similarity graphs is easily ignored in the process of constructing the fusion graph. In addition, the existing method requires an additional clustering step, which causes the similarity graph construction to be separated from the clustering results, so the clustering effect is poor. Summary of the invention
[0005] The present application provides a multi-view clustering method, apparatus and device based on consistency and inconsistency, which is used to solve the technical problem that the prior art ignores the analysis of inconsistency and the relationship analysis between different similarity graphs, and the similarity association analysis is separated from the clustering results, resulting in poor multi-view clustering effect.
[0006] In view of this, the first aspect of the present application provides a multi-view clustering method based on consistency and inconsistency, including:
[0007] The self-expression-based multi-view subspace learning algorithm constructs a similarity graph according to the target view samples and obtains the initial clustering similarity graph matrix;
[0008] By means of matrix decomposition and HSIC dependency analysis, a fusion analysis is performed on the consistent parts and inconsistent parts of the views according to the initial clustering similarity graph matrix to obtain a multi-view fusion graph;
[0009] Cluster analysis is performed on the multi-view fusion graph according to the Laplace rank constraint to obtain a multi-view clustering result.
[0010] Preferably, the self-expression-based multi-view subspace learning algorithm constructs a similarity graph according to the target view samples to obtain an initial clustering similarity graph matrix, including:
[0011] The self-expression-based multi-view subspace learning algorithm constructs a multi-view subspace clustering objective function according to the target view samples;
[0012] Introducing a smoothing normalization term into the multi-view subspace clustering objective function to obtain a self-expressive learning objective function;
[0013] A similarity graph construction operation based on subspace clustering analysis is performed through the self-expression learning objective function to obtain an initial cluster similarity graph matrix.
[0014] Preferably, the method of performing fusion analysis on the consistent parts and inconsistent parts of the views according to the initial clustering similarity graph matrix by means of matrix decomposition and HSIC dependency analysis to obtain a multi-view fusion graph includes:
[0015] Decomposing the consistency part and the inconsistency part from the initial cluster similarity graph matrix by matrix decomposition to obtain a consistency matrix and an inconsistency matrix;
[0016] Using the HSIC standard to perform nonlinear dependency analysis on the inconsistency matrix, and obtaining an inconsistency dependency expression;
[0017] generating a comprehensive view fusion expression by combining the self-expression learning objective function, the inconsistency dependency expression and the consistency expression, wherein the consistency expression is constructed by assigning preset weights to the consistency matrix;
[0018] The view fusion analysis is performed through the comprehensive view fusion expression to obtain a multi-view fusion graph.
[0019] Preferably, performing cluster analysis on the multi-view fusion graph according to the Laplace rank constraint to obtain a multi-view clustering result includes:
[0020] Introducing a Laplace rank constraint into the comprehensive view fusion expression to generate a multi-view clustering objective function;
[0021] A cluster analysis is performed on the multi-view fusion graph according to the multi-view clustering objective function to obtain a multi-view clustering result.
[0022] Preferably, the performing cluster analysis on the multi-view fusion graph according to the Laplace rank constraint to obtain a multi-view clustering result further includes:
[0023] The multi-view clustering analysis process is optimized by means of matrix update and iterative optimization to obtain the optimized clustering result.
[0024] The second aspect of the present application provides a multi-view clustering device based on consistency and inconsistency, including:
[0025] An initial similarity clustering unit is used to construct a similarity graph according to target view samples based on a self-expressive multi-view subspace learning algorithm to obtain an initial clustering similarity graph matrix;
[0026] A multi-view fusion analysis unit, configured to perform fusion analysis on the consistent part and the inconsistent part of the view according to the initial clustering similarity graph matrix by means of matrix decomposition and HSIC dependency analysis, so as to obtain a multi-view fusion graph;
[0027] The multi-view clustering analysis unit is used to perform clustering analysis on the multi-view fusion graph according to the Laplace rank constraint to obtain a multi-view clustering result.
[0028] Preferably, the initial similarity clustering unit is specifically used for:
[0029] The self-expression-based multi-view subspace learning algorithm constructs a multi-view subspace clustering objective function according to the target view samples;
[0030] Introducing a smoothing normalization term into the multi-view subspace clustering objective function to obtain a self-expressive learning objective function;
[0031] A similarity graph construction operation based on subspace clustering analysis is performed through the self-expression learning objective function to obtain an initial cluster similarity graph matrix.
[0032] Preferably, the multi-view fusion analysis unit is specifically used for:
[0033] Decomposing the consistency part and the inconsistency part from the initial cluster similarity graph matrix by matrix decomposition to obtain a consistency matrix and an inconsistency matrix;
[0034] Using the HSIC standard to perform nonlinear dependency analysis on the inconsistency matrix, and obtaining an inconsistency dependency expression;
[0035] generating a comprehensive view fusion expression by combining the self-expression learning objective function, the inconsistency dependency expression and the consistency expression, wherein the consistency expression is constructed by assigning preset weights to the consistency matrix;
[0036] The view fusion analysis is performed through the comprehensive view fusion expression to obtain a multi-view fusion graph.
[0037] Preferably, the multi-view clustering analysis unit is specifically used for:
[0038] Introducing a Laplace rank constraint into the comprehensive view fusion expression to generate a multi-view clustering objective function;
[0039] A cluster analysis is performed on the multi-view fusion graph according to the multi-view clustering objective function to obtain a multi-view clustering result.
[0040] A third aspect of the present application provides a multi-view clustering device based on consistency and inconsistency, the device comprising a processor and a memory;
[0041] The memory is used to store program code and transmit the program code to the processor;
[0042] The processor is used to execute the multi-view clustering method based on consistency and inconsistency according to the instructions in the program code.
[0043] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:
[0044] In the present application, a multi-view clustering method based on consistency and inconsistency is provided, including: constructing a similarity graph according to the target view sample based on a self-expressive multi-view subspace learning algorithm to obtain an initial clustering similarity graph matrix; performing a fusion analysis on the consistent part and the inconsistent part of the view according to the initial clustering similarity graph matrix by means of matrix decomposition and HSIC dependency analysis to obtain a multi-view fusion graph; performing a clustering analysis on the multi-view fusion graph according to the Laplace rank constraint to obtain a multi-view clustering result.
[0045] The multi-view clustering method based on consistency and inconsistency provided by the present application adopts a multi-view subspace learning method based on self-expression to analyze the similarity between sample data, and can preliminarily cluster similar or dissimilar samples. This process can effectively remove noise and anomalies in the data; moreover, the HSIC standard is used to perform dependency analysis on the consistency of views, which can meet the problem of ignoring inconsistency analysis in actual multi-view clustering; in addition, the Laplace rank constraint is used to associate relationship analysis with clustering, which can ensure the accuracy and reliability of clustering results. Therefore, the present application can solve the technical problem that the prior art ignores the analysis of inconsistency and the relationship analysis between different similarity graphs, and the similarity association analysis is separated from the clustering results, resulting in poor multi-view clustering effect. BRIEF DESCRIPTION OF THE DRAWINGS
[0046] Figure 1 A schematic diagram of a multi-view clustering method based on consistency and inconsistency provided in an embodiment of the present application;
[0047] Figure 2 A schematic diagram of the structure of a multi-view clustering device based on consistency and inconsistency provided in an embodiment of the present application;
[0048] Figure 3 An example diagram of movie classification application in the film and television industry provided for the application example of this application. DETAILED DESCRIPTION
[0049] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0050] For easier understanding, see Figure 1 , an embodiment of the multi-view clustering method based on consistency and inconsistency provided by the present application includes:
[0051] Step 101: A self-expression-based multi-view subspace learning algorithm constructs a similarity graph according to target view samples to obtain an initial clustering similarity graph matrix.
[0052] Furthermore, step 101 includes:
[0053] The self-expression-based multi-view subspace learning algorithm constructs a multi-view subspace clustering objective function according to the target view samples;
[0054] A smoothing normalization term is introduced into the multi-view subspace clustering objective function to obtain a self-expressive learning objective function;
[0055] The similarity graph construction operation based on subspace clustering analysis is performed through the self-expression learning objective function to obtain the initial cluster similarity graph matrix.
[0056] It should be noted that if represents the target view sample, represents the i-th attribute of the sample, represents the subspace representation matrix, is the error matrix, then the multi-view analysis based on self-expressive multi-view subspace learning can construct the following function:
[0057]
[0058] Then the constructed multi-view subspace clustering objective function can be expressed as:
[0059]
[0060]
[0061] in, Indicates the number of views. represents the consensus regularization function, which can obtain a global unified graph by fusing each view; Represents the i-th initial cluster similarity graph matrix S i ; represents the square of the norm of the matrix F; 1 is a vector whose elements are all 1.
[0062] If a smoothing normalization term is introduced into the multi-view subspace clustering objective function, the multi-view subspace clustering objective function can be transformed into a self-expressive learning objective function:
[0063]
[0064]
[0065] in, is the smoothing normalization term, represents the trace of the matrix, is the Laplace matrix, is a symmetric matrix and satisfies the equation:
[0066]
[0067] in, is a symmetric matrix The elements in is the adjacency matrix Wi The elements in .
[0068] W i represents the adjacency matrix, which can be constructed using the Gaussian kernel function, defined as:
[0069]
[0070] represents the Gaussian kernel width, , Respectively represent the vectors in the matrix X, w kj is the adjacency matrix W i In order to make the adjacency matrix W i Symmetry can be achieved by the following equation:
[0071]
[0072] in, , Represents the matrix W i The vector in .
[0073] Step 102: By means of matrix decomposition and HSIC dependency analysis, a fusion analysis is performed on the consistent parts and inconsistent parts of the views according to the initial clustering similarity graph matrix to obtain a multi-view fusion graph.
[0074] Further, step 102 includes:
[0075] Decomposing the consistency part and the inconsistency part from the initial cluster similarity graph matrix by matrix decomposition to obtain the consistency matrix and the inconsistency matrix;
[0076] The HSIC standard is used to perform nonlinear dependency analysis on the inconsistency matrix and obtain the inconsistency dependency expression;
[0077] A comprehensive view fusion expression is generated by combining the self-expression learning objective function, the inconsistent dependency expression and the consistent expression. The consistent expression is constructed by assigning preset weights to the consistency matrix.
[0078] The view fusion analysis is performed through the comprehensive view fusion expression to obtain the multi-view fusion graph.
[0079] It should be noted that the consistency part and the inconsistency part of the view can be decomposed in the similarity graph matrix to obtain the consistency matrix A i and the inconsistency matrix D i , which can be understood as the i-th similarity graph matrix S in multiple views i It is divided into two parts, specifically expressed as:
[0080]
[0081] So, the similarity graph matrix S i Also called a similarity matrix. In order to integrate different similarity matrices into a unified consensus similarity graph, it is necessary to consider fusing different similarity graphs. In this embodiment, a preset weight can be assigned to the consistency matrix first, and then integrated into a common similarity graph to obtain a consistency expression:
[0082]
[0083]
[0084] in, represents the weight vector, Represents the fusion graph matrix.
[0085] In order to make the inconsistent parts in the view diversified and independent of each other, so as to enhance the complementarity of information, this embodiment uses the Hilbert-Schmidt Independence Criterion (HSIC) standard to perform nonlinear dependency analysis on the inconsistency matrix to obtain the inconsistency dependency expression; it is mainly based on this standard to measure the dependency of the inconsistent parts in the view. HSIC uses the reconstructed kernel Hibbert space to measure the correlation between variable mappings, which can solve the nonlinear dependency problem. The obtained inconsistency dependency expression is:
[0086]
[0087] in, represents the Hilbert-Schmidt independence criterion function, H is the centering matrix, , .
[0088] The combined self-expression learning objective function, inconsistent dependency expression and consistent expression can generate a comprehensive view fusion expression:
[0089]
[0090]
[0091]
[0092]
[0093] in, , , All of them are parameters. Through the comprehensive view fusion expression, a multi-view fusion analysis is performed on multiple initial clustering similarity graphs to obtain a multi-view fusion graph, which is an element in the vector.
[0094] Step 103: Perform cluster analysis on the multi-view fusion graph according to the Laplace rank constraint to obtain a multi-view clustering result.
[0095] Furthermore, step 103 includes:
[0096] The Laplace rank constraint is introduced into the comprehensive view fusion expression to generate the multi-view clustering objective function;
[0097] The multi-view fusion graph is clustered according to the multi-view clustering objective function to obtain the multi-view clustering result.
[0098] It should be noted that the theorem of the Laplace rank constraint is expressed as: The Laplace matrix L u The multiplicity of eigenvalue 0 is equal to the number of graph-connected components of the matrix U. Assume YesL u The i-th smallest eigenvalue, and ,if , then , where n is the number of samples. The minimum eigenvalue of the solution is:
[0099]
[0100] Among them, Embedding matrix. That is, the Laplace matrix L u The eigenvectors of , and I is the identity matrix.
[0101] Based on the above, we can see that by introducing the Laplace rank constraint into the comprehensive view fusion expression, the generated multi-view clustering objective function is expressed as:
[0102]
[0103]
[0104]
[0105]
[0106] in, Also a parameter.
[0107] It should be noted that this embodiment imposes a Laplace rank constraint on the fusion graph matrix, and can directly obtain the clustering result after solving the calculation without additional clustering operation. At the same time, this operation makes the construction of the similarity graph and the final clustering result have a certain correlation, which can improve the accuracy of classification.
[0108] Furthermore, step 103 further includes:
[0109] The multi-view clustering analysis process is optimized by means of matrix update and iterative optimization to obtain the optimized clustering result.
[0110] It should be noted that in order to obtain more accurate optimized clustering results, this embodiment can also update and iteratively optimize the above clustering process. By fixing certain matrix parameters in the multi-view clustering objective function and updating a certain matrix or parameter, the clustering solution process is continuously optimized until a stable and accurate clustering result is obtained.
[0111] Specifically, you can fix A i ,U,D i , F and , update S i ; The operation is to remove irrelevant items from the initial clustering similarity graph matrix and retain relevant items, then the following optimization function can be obtained:
[0112]
[0113]
[0114] The above formula can be further simplified:
[0115]
[0116]
[0117] in, is the Lagrange multiplier, is the penalty coefficient, let
[0118]
[0119] At the same time, for S in the above formula i Find the partial derivative and set it equal to 0, and you can get S i The optimal solution update function of :
[0120]
[0121] Among them, L i is the Laplace matrix, , is the penalty coefficient. It can be seen that the above equation satisfies the standard Sylvester equation and has a unique solution; in order to satisfy the constraint S i , let S i =max(S i , 0).
[0122] You can also fix S i ,U,D i , F and , update A i ; Remove and A i The irrelevant terms of , retaining the relevant terms, can be obtained as follows:
[0123]
[0124]
[0125] The above formula can be solved using the enhanced Lagrangian function to obtain:
[0126]
[0127]
[0128] For the above formula, A i By taking the partial derivative of and setting it equal to 0, we can get A i The optimal solution update function of :
[0129]
[0130] Similarly, in order to satisfy constraint A i , let A i =max(A i , 0).
[0131] You can also fix A i ,U,S i , F and , update D i ; Remove and D i The irrelevant terms of , retaining the relevant terms, can obtain the following optimization function:
[0132]
[0133]
[0134] Using the inner product kernel formula for HSIC, the above formula can be changed to:
[0135]
[0136]
[0137] in,
[0138]
[0139] The above formula is optimized by using the conditional Lagrangian function, and the optimization function can be equivalent to:
[0140]
[0141] For the above formula, D i Find the partial derivative of and set the partial derivative equal to 0, then we get D i The optimal update function can be expressed as:
[0142]
[0143] Where I is the identity matrix.
[0144] You can also fix A i , S i , D i , F and , update U; its optimization objective function is:
[0145]
[0146]
[0147] According to the properties of matrix trace, the above formula can be simplified as:
[0148]
[0149]
[0150] in, , denote the vectors of the embedding matrix, Represents the i-th matrix A i The element in row j and column k of Represents the element in the jth row and kth column of the matrix U, and is defined as:
[0151]
[0152] For different j, U is independent, so we can get the following formula:
[0153]
[0154]
[0155] in, , , They are respectively the vector of the i-th view A, the vector of the fusion matrix U, and the vector of the inconsistency matrix D. Redefine:
[0156]
[0157] Then the above formula can be simplified to:
[0158]
[0159]
[0160] Solving the above formula, we can get the updated U.
[0161] You can also fix A i ,U,D i , F and S i ,renew ; Its objective function is:
[0162]
[0163]
[0164] Its Lagrangian function is expressed as:
[0165]
[0166] in, is the Lagrange multiplier, in the above formula Find the partial derivative of and make it equal to 0:
[0167]
[0168] in, represents the Lagrange coefficient vector.
[0169] Further simplification yields:
[0170]
[0171] Therefore, we can solve:
[0172]
[0173] You can also fix A i ,U,D i , S i and , update F; F is determined by L u of The minimum eigenvalue corresponds to The optimization objective function is expressed as:
[0174]
[0175]
[0176] For ease of understanding, this application takes the film classification of the film and television industry as an example, please refer to Figure 3, and uses a clustering algorithm to classify movies by analyzing the attributes of movies such as type, director, actor, comment, duration, movie color, language, year, etc. 28 information, so that users can quickly find the movie category of interest according to their preferences, thereby realizing personalized recommendations and improving user satisfaction with watching movies. In the movie recommendation system, the original data usually contains noise, outliers and outliers. For example, a movie may be regarded as noise or outliers because of factors such as incorrect description of the name of the director or actor or excessive difference in ratings on different platforms. Traditional clustering algorithms directly use these original data without excluding the influence of these outliers and noise, resulting in a decrease in the quality of each view in the multi-view clustering method. In addition, different movie attributes have different degrees of influence on movie classification. Among these attributes, the category has the greatest impact on movie classification. However, existing methods often treat each viewing attribute equally, without considering their differences in importance, and simply fuse the data graphs of different view attributes. This approach may cause the final fusion graph to deviate greatly from the actual situation.
[0177] If the multi-view clustering method based on consistency and inconsistency of the present application is used to classify movies, the similarity between movie sample data can be learned based on the self-expressive multi-view subspace learning method to construct a reliable similarity graph; the fusion of the consistent and inconsistent parts can focus on the dependency of the inconsistency to ensure the accuracy of the fusion graph; finally, the Laplace rank constraint can be used to associate the multi-view analysis with the clustering results, thereby improving the reliability of the clustering results.
[0178] The multi-view clustering method based on consistency and inconsistency provided in the embodiment of the present application adopts a multi-view subspace learning method based on self-expression to analyze the similarity between sample data, and can preliminarily cluster similar or dissimilar samples. This process can effectively remove noise and anomalies in the data; moreover, the HSIC standard is used to perform dependency analysis on the consistency of views, which can meet the problem of ignoring inconsistency analysis in actual multi-view clustering; in addition, the Laplace rank constraint is used to associate relationship analysis with clustering, which can ensure the accuracy and reliability of clustering results. Therefore, the embodiment of the present application can solve the technical problem that the prior art ignores the analysis of inconsistency and the relationship analysis between different similarity graphs, and the similarity association analysis is separated from the clustering results, resulting in poor multi-view clustering effect.
[0179] For easier understanding, see Figure 2 The present application provides an embodiment of a multi-view clustering device based on consistency and inconsistency, including:
[0180] An initial similarity clustering unit 201 is used to construct a similarity graph according to target view samples based on a self-expressive multi-view subspace learning algorithm to obtain an initial clustering similarity graph matrix;
[0181] The multi-view fusion analysis unit 202 is used to perform fusion analysis on the consistent part and the inconsistent part of the view according to the initial clustering similarity graph matrix by means of matrix decomposition and HSIC dependency analysis to obtain a multi-view fusion graph;
[0182] The multi-view clustering analysis unit 203 is used to perform clustering analysis on the multi-view fusion graph according to the Laplace rank constraint to obtain a multi-view clustering result.
[0183] Furthermore, the initial similarity clustering unit 201 is specifically used for:
[0184] The self-expression-based multi-view subspace learning algorithm constructs a multi-view subspace clustering objective function according to the target view samples;
[0185] A smoothing normalization term is introduced into the multi-view subspace clustering objective function to obtain a self-expressive learning objective function;
[0186] The similarity graph construction operation based on subspace clustering analysis is performed through the self-expression learning objective function to obtain the initial cluster similarity graph matrix.
[0187] Furthermore, the multi-view fusion analysis unit 202 is specifically configured to:
[0188] Decomposing the consistency part and the inconsistency part from the initial cluster similarity graph matrix by matrix decomposition to obtain the consistency matrix and the inconsistency matrix;
[0189] The HSIC standard is used to perform nonlinear dependency analysis on the inconsistency matrix and obtain the inconsistency dependency expression;
[0190] A comprehensive view fusion expression is generated by combining the self-expression learning objective function, the inconsistent dependency expression and the consistent expression. The consistent expression is constructed by assigning preset weights to the consistency matrix.
[0191] The view fusion analysis is performed through the comprehensive view fusion expression to obtain the multi-view fusion graph.
[0192] Furthermore, the multi-view clustering analysis unit 203 is specifically configured to:
[0193] The Laplace rank constraint is introduced into the comprehensive view fusion expression to generate the multi-view clustering objective function;
[0194] The multi-view fusion graph is clustered according to the multi-view clustering objective function to obtain the multi-view clustering result.
[0195] The present application also provides a multi-view clustering device based on consistency and inconsistency, the device comprising a processor and a memory;
[0196] The memory is used to store the program code and transmit the program code to the processor;
[0197] The processor is used to execute the multi-view clustering method based on consistency and inconsistency in the above method embodiment according to the instructions in the program code.
[0198] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.
[0199] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0200] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.
[0201] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a number of instructions for executing all or part of the steps of the method described in each embodiment of the present application through a computer device (which can be a personal computer, server, or network device, etc.). The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (full name in English: Read-Only Memory, English abbreviation: ROM), random access memory (full name in English: Random Access Memory, English abbreviation: RAM), disk or optical disk and other media that can store program codes.
[0202] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.
Claims
1. A multi-view clustering method based on consistency and inconsistency, characterized by: include: The self-expression-based multi-view subspace learning algorithm constructs a similarity graph according to the target view samples and obtains the initial clustering similarity graph matrix; By means of matrix decomposition and HSIC dependency analysis, a fusion analysis is performed on the consistent parts and inconsistent parts of the views according to the initial clustering similarity graph matrix to obtain a multi-view fusion graph; Cluster analysis is performed on the multi-view fusion graph according to the Laplace rank constraint to obtain a multi-view clustering result.
2. The multi-view clustering method based on consistency and inconsistency according to claim 1, characterized in that: The self-expression-based multi-view subspace learning algorithm constructs a similarity graph according to the target view samples to obtain an initial clustering similarity graph matrix, including: The self-expression-based multi-view subspace learning algorithm constructs a multi-view subspace clustering objective function according to the target view samples; Introducing a smoothing normalization term into the multi-view subspace clustering objective function to obtain a self-expressive learning objective function; A similarity graph construction operation based on subspace clustering analysis is performed through the self-expression learning objective function to obtain an initial cluster similarity graph matrix.
3. The multi-view clustering method based on consistency and inconsistency according to claim 2, characterized in that: The method of performing fusion analysis on the consistent part and the inconsistent part of the view according to the initial clustering similarity graph matrix by means of matrix decomposition and HSIC dependency analysis to obtain a multi-view fusion graph includes: Decomposing the consistency part and the inconsistency part from the initial cluster similarity graph matrix by matrix decomposition to obtain a consistency matrix and an inconsistency matrix; Using the HSIC standard to perform nonlinear dependency analysis on the inconsistency matrix, and obtaining an inconsistency dependency expression; generating a comprehensive view fusion expression by combining the self-expression learning objective function, the inconsistency dependency expression and the consistency expression, wherein the consistency expression is constructed by assigning preset weights to the consistency matrix; The view fusion analysis is performed through the comprehensive view fusion expression to obtain a multi-view fusion graph.
4. The multi-view clustering method based on consistency and inconsistency according to claim 3, characterized in that: The clustering analysis is performed on the multi-view fusion graph according to the Laplace rank constraint to obtain a multi-view clustering result, including: Introducing a Laplace rank constraint into the comprehensive view fusion expression to generate a multi-view clustering objective function; A cluster analysis is performed on the multi-view fusion graph according to the multi-view clustering objective function to obtain a multi-view clustering result.
5. The multi-view clustering method based on consistency and inconsistency according to claim 1, characterized in that: The clustering analysis is performed on the multi-view fusion graph according to the Laplace rank constraint to obtain a multi-view clustering result, and further includes: The multi-view clustering analysis process is optimized by means of matrix update and iterative optimization to obtain the optimized clustering result.
6. A multi-view clustering device based on consistency and inconsistency, characterized in that: include: An initial similarity clustering unit is used to construct a similarity graph according to target view samples based on a self-expressive multi-view subspace learning algorithm to obtain an initial clustering similarity graph matrix; A multi-view fusion analysis unit, configured to perform fusion analysis on the consistent part and the inconsistent part of the view according to the initial clustering similarity graph matrix by means of matrix decomposition and HSIC dependency analysis, so as to obtain a multi-view fusion graph; The multi-view clustering analysis unit is used to perform clustering analysis on the multi-view fusion graph according to the Laplace rank constraint to obtain a multi-view clustering result.
7. The multi-view clustering device based on consistency and inconsistency according to claim 6, characterized in that: The initial similarity clustering unit is specifically used for: The self-expression-based multi-view subspace learning algorithm constructs a multi-view subspace clustering objective function according to the target view samples; Introducing a smoothing normalization term into the multi-view subspace clustering objective function to obtain a self-expressive learning objective function; A similarity graph construction operation based on subspace clustering analysis is performed through the self-expression learning objective function to obtain an initial cluster similarity graph matrix.
8. The multi-view clustering device based on consistency and inconsistency according to claim 7, characterized in that: The multi-view fusion analysis unit is specifically used for: Decomposing the consistency part and the inconsistency part from the initial cluster similarity graph matrix by matrix decomposition to obtain a consistency matrix and an inconsistency matrix; Using the HSIC standard to perform nonlinear dependency analysis on the inconsistency matrix, and obtaining an inconsistency dependency expression; generating a comprehensive view fusion expression by combining the self-expression learning objective function, the inconsistency dependency expression and the consistency expression, wherein the consistency expression is constructed by assigning preset weights to the consistency matrix; The view fusion analysis is performed through the comprehensive view fusion expression to obtain a multi-view fusion graph.
9. The multi-view clustering device based on consistency and inconsistency according to claim 8, characterized in that: The multi-view clustering analysis unit is specifically used for: Introducing a Laplace rank constraint into the comprehensive view fusion expression to generate a multi-view clustering objective function; A cluster analysis is performed on the multi-view fusion graph according to the multi-view clustering objective function to obtain a multi-view clustering result.
10. A multi-view clustering device based on consistency and inconsistency, characterized in that The device comprises a processor and a memory; The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the consistency and inconsistency based multi-view clustering method according to any one of claims 1 to 5 according to the instructions in the program code.
