Multi-view data clustering method and device, electronic equipment and storage medium
By performing data reconstruction and low-rank constraints on multi-view data, combining self-representation learning, regularization, independence processing and adaptive graph learning terms, the final data matrix is generated, which solves the problem of poor clustering effect in the existing technology and realizes more effective multi-view data clustering.
Patent Information
- Application Number
- CN202411743607.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-30
- Publication Date
- 2025-02-25
AI Technical Summary
The existing multi-view data clustering method cannot effectively deal with the data structural damage caused by noise and outlier pollution, resulting in poor clustering effect.
By reconstructing the multi-view data in different subspaces and adding low-rank constraints, the affinity matrix is generated, and regularization and independence are performed, and finally the final data matrix is generated using the adaptive graph learning term.
This method can fully tap the potential structural information of multi-view data, improve clustering effect, and reduce the impact of noise and outliers.
Smart Images
Figure CN119516230A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of data processing technology, and in particular to a multi-view data clustering method, device, electronic device and storage medium. Background Art
[0002] In the multi-view clustering model, how to fully mine the information contained in multiple views and avoid the influence of noise is the key to establishing an effective learning model.
[0003] However, most of the existing multi-view data clustering methods directly use the original data for clustering, which cannot deal with the structural damage of data caused by noise and outlier pollution, resulting in poor clustering effect. Summary of the invention
[0004] In view of this, the present application provides a multi-view data clustering method, device, electronic device and storage medium to solve the problem of poor clustering effect in the prior art.
[0005] To achieve the above objectives, this application provides the following technical solutions:
[0006] The first aspect of the present application discloses a multi-view data clustering method, comprising:
[0007] For the multi-view data, reconstruct the data in each view in different subspaces to obtain a data matrix of the multi-view data in each view;
[0008] For each view, adding a low-rank constraint to the data matrix to obtain a low-rank data matrix;
[0009] For each low-rank data matrix corresponding to each view, the low-rank data matrix corresponding to each view is processed by a self-representation learning method to generate an affinity matrix of the multi-view data;
[0010] Regularizing the affinity matrix of the multi-view data to obtain a processed affinity matrix;
[0011] Performing independence processing on the processed affinity matrix by using the Hilbert-Schmidt independence criterion to obtain an independent affinity matrix;
[0012] The independent affinity matrix is processed using an adaptive graph learning term to obtain a final data matrix corresponding to the multi-view data.
[0013] Optionally, the above method, before reconstructing the data in each view in different subspaces to obtain the data matrix of the multi-view data in each view for the multi-view data, further comprises:
[0014] Data preprocessing is performed on the multi-view data to obtain processed multi-view data.
[0015] Optionally, the above method, after performing data preprocessing on the multi-view data to obtain processed multi-view data, further comprises:
[0016] An original data matrix of the processed multi-view data in each view is generated.
[0017] Optionally, the above method, after using the adaptive graph learning item to process the independent affinity matrix to obtain a final data matrix corresponding to the multi-view data, further includes:
[0018] Based on the final data matrix, cluster labels of the multi-view data are generated.
[0019] The second aspect of the present application discloses a multi-view data clustering device, comprising:
[0020] A reconstruction unit, configured to reconstruct the data in each view in different subspaces for the multi-view data, so as to obtain a data matrix of the multi-view data in each view;
[0021] A constraint unit, used for adding a low-rank constraint to the data matrix for each view to obtain a low-rank data matrix;
[0022] A first processing unit is configured to process the low-rank data matrix corresponding to each view by a self-representation learning method to generate an affinity matrix of the multi-view data;
[0023] A second processing unit is used to perform regularization processing on the affinity matrix of the multi-view data to obtain a processed affinity matrix;
[0024] A third processing unit is used to perform independence processing on the processed affinity matrix by using the Hilbert-Schmidt independence criterion to obtain an independent affinity matrix;
[0025] The fourth processing unit is used to process the independent affinity matrix using the adaptive graph learning item to obtain a final data matrix corresponding to the multi-view data.
[0026] Optionally, the above device further includes:
[0027] The preprocessing unit is used to perform data preprocessing on the multi-view data to obtain processed multi-view data.
[0028] Optionally, the above device further includes:
[0029] The first generating unit is used to generate an original data matrix of the processed multi-view data in each view.
[0030] Optionally, the above device further includes:
[0031] The second generating unit is configured to generate cluster labels for the multi-view data based on the final data matrix.
[0032] The third aspect of the present application discloses an electronic device, comprising:
[0033] one or more processors;
[0034] a storage device having one or more programs stored thereon;
[0035] When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of the first aspects of the present invention.
[0036] A fourth aspect of the present application discloses a computer storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the method described in any one of the first aspects of the present invention is implemented.
[0037] It can be seen from the above technical scheme that in a clustering method for multi-view data provided by the present application, first, for the multi-view data, the data under each view is reconstructed in different subspaces to obtain the data matrix of the multi-view data under each view. Then, for each view, a low-rank constraint is added to the data matrix to obtain a low-rank data matrix. Then, for the low-rank data matrix corresponding to each view, the low-rank data matrix corresponding to each view is processed by a self-representation learning method to generate an affinity matrix of the multi-view data. The affinity matrix of the multi-view data is then regularized to obtain a processed affinity matrix. The processed affinity matrix is subjected to independence processing by the Hilbert-Schmidt independence criterion to obtain an independent affinity matrix. Finally, the independent affinity matrix is processed using an adaptive graph learning item to obtain the final data matrix corresponding to the multi-view data. It can be seen that the method of the present application takes into account the high dimensionality and complexity of the original data, reconstructs the data under each view in different subspaces, adds low-rank constraints, and then generates an affinity matrix of the multi-view data. It is processed by combining self-representation learning, Hilbert-Schmidt independence criterion and adaptive graph learning terms, which can fully explore the potential structural information of each view and improve the clustering effect. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.
[0039] Figure 1 A flowchart of a multi-view data clustering method disclosed in an embodiment of the present application;
[0040] Figure 2 A schematic diagram of a multi-view data clustering device disclosed in an embodiment of the present application;
[0041] Figure 3 A schematic diagram of an electronic device disclosed in an embodiment of the present application. DETAILED DESCRIPTION
[0042] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions 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.
[0043] In this application, the terms "comprises", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element.
[0044] Furthermore, in this document, relational terms such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, but do not necessarily require or imply any such actual relationship or order between these entities or operations.
[0045] Multi-view clustering can make full use of the consistency and independence information between views and fully explore the potential structural information of multi-view data. It has become an increasingly common research topic and is widely used in data mining, cognitive computing and other fields.
[0046] However, many existing multi-view clustering algorithms have some problems. First, many of them directly process the original data space using techniques such as self-representation and non-negative matrix factorization. In such a space, the prevalence of high-dimensional and redundant features may lead to dimensionality fraud (in order to improve performance, some features or dimensions are added or modified, but these features or dimensions are irrelevant to the real world or should not be considered, which leads to poor performance of the model in practical applications) and dimensionality curse (in high-dimensional space, data sparsity and dimensionality disaster problems will cause the model to overfit or fail, thereby reducing the performance and reliability of the model), which makes the model calculation complex and reduces performance. In addition, noise and outliers in the original space will destroy the underlying clustering structure of the data, thereby generating unreliable and inaccurate clustering results. Failure to fully consider the consistency and independence of each view in fusion will also cause effective information to be lost.
[0047] As can be seen from the background technology, in a multi-view clustering model, how to fully exploit the information contained in multiple views and avoid the influence of noise is the key to establishing an effective learning model. However, most existing multi-view data clustering methods directly use the original data for clustering, which cannot deal well with the structural damage of data caused by noise and outlier pollution, resulting in poor clustering effect.
[0048] In view of this, the present application provides a multi-view data clustering method, device, electronic device and storage medium to solve the problem of poor clustering effect in the prior art.
[0049] The present application embodiment provides a multi-view data clustering method, such as Figure 1 As shown, specifically including:
[0050] S101 . For multi-view data, reconstruct the data in each view in different subspaces to obtain a data matrix of the multi-view data in each view.
[0051] It should be noted that multi-view data means that the data has multiple forms or consists of multiple representations. For face data, it can be described from multiple angles; multimedia data consists of sound, images, and text. Multi-view data describes the characteristics of things from multiple aspects, so it contains more information than single-view data. From the final performance point of view, multi-view data means a data matrix with multiple samples.
[0052] It should also be noted that the existence of high-dimensional and redundant features in the original space as well as the existence of noise and outliers leads to the unreliability of the original data. Therefore, in order to effectively consider the independence of multi-view data, for multi-view data, the data under each view is reconstructed in different subspaces to obtain the data matrix of the multi-view data under each view. The data reconstruction expression is as follows:
[0053] X (v) =W (v) H (v) +E (v)
[0054] Among them, the above are all data under view v, X (v) refers to the original data matrix, W (v) Refers to the projection matrix, H (v) refers to the low-dimensional reconstructed data matrix, E (v) is the error matrix.
[0055] Optionally, in another embodiment of the present application, before executing step S101, the following may also be included:
[0056] The multi-view data is preprocessed to obtain processed multi-view data.
[0057] It should be noted that in order to remove redundant data and reduce noise interference, data preprocessing such as data cleaning and refining is performed on the multi-view data to obtain processed multi-view data.
[0058] Optionally, in another embodiment of the present application, after performing the above steps to pre-process the multi-view data to obtain the processed multi-view data, the following may also be included:
[0059] Generate the original data matrix of the processed multi-view data in each view.
[0060] It should be noted that, after data preprocessing, processed multi-view data is obtained, and then an original data matrix corresponding to each view is generated in the original space according to the processed multi-view data, so as to perform subsequent data reconstruction.
[0061] S102. For each view, add a low-rank constraint to the data matrix to obtain a low-rank data matrix.
[0062] It should be noted that since assigning a suitable dimension value to each view is a difficult problem, the projection dimensions of all views need to be unified. Low-rank constraints have a certain promoting effect on the sparsity and dimensionality reduction of the matrix. Therefore, for each view, a low-rank constraint is added to the data matrix to obtain low-rank data corresponding to each view, so as to improve the robustness of the model and suppress the negative impact of forced consistent dimensional projection. In addition, since the noise of cluster points will increase the rank of the data, high-rank data usually has a corresponding low-rank structure. Therefore, such a rank constraint can ensure the cleanliness of the reconstructed matrix to a certain extent. Therefore, there is:
[0063]
[0064] xT(v) =W (v) H (v) +E (v) ,W (v) (W (v) ) T =I
[0065] Among them, λ1 is the weight coefficient and I is the unit matrix.
[0066] S103, for each low-rank data matrix corresponding to each view, processing the low-rank data matrix corresponding to each view by a self-representation learning method to generate an affinity matrix of multi-view data;
[0067] It should be noted that after obtaining the low-rank data matrix of each view, the low-rank data matrix corresponding to each view is processed by the self-representation learning method to generate an affinity matrix for multi-view data. Compared with adaptive graph learning, self-representation learning is more comprehensive and easier to consider the relationship between views, which is conducive to the subsequent consideration of view independence. Among them, the affinity matrix is a data matrix that can eventually be processed by general clustering methods, and the value of each element represents the weight coefficient of each dimension. The goal of the multi-view clustering model is generally a consistent affinity matrix.
[0068] S104 . Regularize the affinity matrix of the multi-view data to obtain a processed affinity matrix.
[0069] It should be noted that when performing self-representation learning, it is easy to ignore the local structural information of the graph. Therefore, the affinity matrix of the multi-view data is regularized through Laplace matrix regularization to obtain the processed affinity matrix. Laplace regularization can introduce local relationships and better capture the structural characteristics of the graph. By minimizing the difference between the predicted value and the true value, the nodes in the self-representation learning process have similar embedding vectors in the low-dimensional space. The processed affinity matrix takes into account both the global and local structures of the reconstructed data. The processed affinity matrix is as follows:
[0070]
[0071] xT (v) =W (v) H (v) +E (v) ,H (v) =H (v) Z (v) ,W (v) (W (v) ) T =I
[0072] Among them, Z (v) is the coefficient matrix obtained by self-representation learning, and α is its weight.
[0073] Self-representation learning is one of the methods to obtain affinity matrices. Similar methods include adaptive affinity graph learning, matrix decomposition, etc. In general, self-representation learning assumes that each data point in the data set can be approximately linearly represented by other data points, so a coefficient matrix can be obtained, and then the affinity matrix can be obtained.
[0074] S105. Perform independence processing on the processed affinity matrix using the Hilbert-Schmidt independence criterion to obtain an independent affinity matrix.
[0075] It should be noted that the affinity matrix after the above processing does not consider the relationship between views. Therefore, the Hilbert-Schmidt independence criterion is used to perform independence processing on the affinity matrix after the processing to enhance the independence between views and obtain an independent affinity matrix. Then:
[0076]
[0077] xT (v) =W (v) H (v) +E (v) ,H (v) =H (v) Z (v) ,W (v) (W (v) ) T =I
[0078] Among them, λ2 is the parameter of HSIC (Hilbert-Schmidt independence criterion). The Hilbert-Schmidt independence criterion is a measure of statistical dependence between two random variables. After transformation, it can be used to calculate the independence between two matrices. The smaller the value, the higher the independence between the matrices.
[0079] S106. Using adaptive graph learning items to process the independent affinity matrix, obtain a final data matrix corresponding to the multi-view data.
[0080] It should be noted that the independent affinity matrix is processed by adaptive graph learning terms to constrain the affinity matrix of a single view to obtain the final data matrix. Therefore, the proposed final can be expressed as:
[0081]
[0082] xT (v) =W (v) H (v) +E (v) ,H (v) =H (v) Z (v) ,W (v) (W(v) ) T =I
[0083] Among them, L s is the Laplace matrix.
[0084] The adaptive graph learning term can effectively preserve the local geometric structure of the data and reduce the damage to the data structure caused by the Hilbert-Schmidt Independence Criterion (HSIC). In addition, this term ensures the grouping effect of the model in an explicit form, which helps to improve the performance of the clustering model.
[0085] In a clustering method for multi-view data provided in an embodiment of the present application, first, for the multi-view data, the data under each view is reconstructed in different subspaces to obtain the data matrix of the multi-view data under each view. Then, for each view, a low-rank constraint is added to the data matrix to obtain a low-rank data matrix. Then, for the low-rank data matrix corresponding to each view, the low-rank data matrix corresponding to each view is processed by a self-representation learning method to generate an affinity matrix for the multi-view data. The affinity matrix of the multi-view data is then regularized to obtain a processed affinity matrix. The processed affinity matrix is subjected to independence processing by the Hilbert-Schmidt independence criterion to obtain an independent affinity matrix. Finally, the independent affinity matrix is processed using an adaptive graph learning item to obtain the final data matrix corresponding to the multi-view data. It can be seen that the method of the present application takes into account the high dimensionality and complexity of the original data, reconstructs the data under each view in different subspaces, adds low-rank constraints, and then generates an affinity matrix for multi-view data. It combines self-representation learning, Hilbert-Schmidt independence criterion, and adaptive graph learning terms for processing, which can fully explore the potential structural information of each view and improve the clustering effect.
[0086] Optionally, in another embodiment of the present application, after executing step S106, the following may also be included:
[0087] Based on the final data matrix, cluster labels for multi-view data are generated.
[0088] It should be noted that the final data matrix is processed by the standard spectral clustering algorithm to generate cluster labels for the multi-view data. The standard spectral clustering algorithm is a clustering method based on graph theory, which clusters the eigenvectors of the Laplacian matrix of the sample data to achieve the purpose of clustering the sample data.
[0089] Optionally, in another embodiment of the present application, a clustering model is constructed in combination with the above method, and the entire model consists of five parts. The first part performs low-rank constraints on the reconstructed data matrix, with the aim of improving the adaptability of the model to a variety of data and reducing the negative impact of unified dimensionality reduction on the model. The second part applies the 2,1 norm to the error matrix to process the decomposition error of the first part and the sample noise or outliers in the original data. The third part is the regularization term of the coefficient matrix in the self-representation term. The fourth part uses HSIC technology for view fusion, enhances view independence, and ensures full utilization of multi-view information. The last part is an adaptive graph learning item, the purpose of which is to effectively retain the local geometric structure of the data and reduce damage to the data structure in HSIC. In addition, this item ensures the grouping effect of the model in an explicit form, which helps to improve the performance of the clustering model.
[0090] Compared with other multi-view clustering models, this model fully considers the high dimensionality and complexity of the original data, reconstructs the multi-view data in multiple low-dimensional and low-rank spaces, and independently obtains a clean new data matrix. This model combines self-representation learning, HSIC, and adaptive graph learning items to fully explore the potential structural information of each view and utilize the consistency and independence between them.
[0091] The specific representation of the model is as follows:
[0092]
[0093]
[0094] In addition, using ADMM (Alternating Direction Method of Multipliers) to solve the model allows the model to converge quickly within 20 times. The experimental method also proves the effectiveness of the model. ADMM is a computational framework for solving optimization problems, suitable for solving distributed convex optimization problems. It decomposes large global problems into multiple easy-to-solve sub-problems and updates them iteratively.
[0095] Another embodiment of the present application also discloses a multi-view data clustering device, such as Figure 2 As shown, specifically including:
[0096] The reconstruction unit 201 is used to reconstruct the data in each view in different subspaces for the multi-view data to obtain a data matrix of the multi-view data in each view.
[0097] The constraint unit 202 is used to add a low-rank constraint to the data matrix for each view to obtain a low-rank data matrix.
[0098] The first processing unit 203 is used to process the low-rank data matrix corresponding to each view by a self-representation learning method to generate an affinity matrix of multi-view data.
[0099] The second processing unit 204 is configured to perform regularization processing on the affinity matrix of the multi-view data to obtain a processed affinity matrix.
[0100] The third processing unit 205 is used to perform independence processing on the processed affinity matrix by using the Hilbert-Schmidt independence criterion to obtain an independent affinity matrix.
[0101] The fourth processing unit 206 is used to process the independent affinity matrix using the adaptive graph learning item to obtain a final data matrix corresponding to the multi-view data.
[0102] In this embodiment, the specific execution process of the reconstruction unit 201, the constraint unit 202, the first processing unit 203, the second processing unit 204, the third processing unit 205 and the fourth processing unit 206 can be referred to the corresponding Figure 1 The content of the method embodiment will not be repeated here.
[0103] In a multi-view data clustering device provided in an embodiment of the present application, first, the reconstruction unit 201 reconstructs the data under each view in different subspaces for the multi-view data to obtain the data matrix of the multi-view data under each view. Then the constraint unit 202 adds low-rank constraints to the data matrix for each view to obtain a low-rank data matrix. Then the first processing unit 203 processes the low-rank data matrix corresponding to each view by a self-representation learning method to generate an affinity matrix for the multi-view data. The second processing unit 204 then regularizes the affinity matrix of the multi-view data to obtain the processed affinity matrix. The third processing unit 205 performs independence processing on the processed affinity matrix by the Hilbert-Schmidt independence criterion to obtain an independent affinity matrix. Finally, the fourth processing unit 206 processes the independent affinity matrix using an adaptive graph learning item to obtain the final data matrix corresponding to the multi-view data. It can be seen that the device of the present application takes into account the high dimensionality and complexity of the original data, reconstructs the data under each view in different subspaces, adds low-rank constraints, and then generates an affinity matrix for multi-view data. It combines self-representation learning, Hilbert-Schmidt independence criterion, and adaptive graph learning terms for processing, which can fully explore the potential structural information of each view and improve the clustering effect.
[0104] Optionally, in another embodiment of the present application, the multi-view data clustering device may further include:
[0105] The preprocessing unit is used to perform data preprocessing on the multi-view data to obtain processed multi-view data.
[0106] In this embodiment, the specific execution process of the preprocessing unit can refer to the corresponding method embodiment content mentioned above, which will not be repeated here.
[0107] Optionally, in another embodiment of the present application, the multi-view data clustering device may further include:
[0108] The first generating unit is used to generate an original data matrix of the processed multi-view data in each view.
[0109] In this embodiment, the specific execution process of the first generating unit can refer to the corresponding method embodiment content mentioned above, which will not be repeated here.
[0110] The second generating unit is used to generate clustering labels of the multi-view data based on the final data matrix.
[0111] In this embodiment, the specific execution process of the second generation unit can refer to the corresponding method embodiment content mentioned above, which will not be repeated here.
[0112] Another embodiment of the present application further provides an electronic device, such as Figure 3 As shown, specifically including:
[0113] One or more processors 301 .
[0114] The storage device 302 stores one or more programs.
[0115] When one or more programs are executed by one or more processors 301 , the one or more processors 301 implement any one of the methods in the above embodiments.
[0116] Another embodiment of the present application further provides a computer storage medium on which a computer program is stored, wherein when the computer program is executed by a processor, any one of the methods in the above embodiments is implemented.
[0117] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, and the computer program contains a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device, or installed from a ROM. When the computer program is executed by a processing device, the above-mentioned functions defined in the method of the embodiment of the present disclosure are executed.
[0118] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can refer to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiment. The system and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without creative work.
[0119] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.
[0120] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A clustering method for multi-view data, characterized in that: include: For the multi-view data, reconstruct the data in each view in different subspaces to obtain a data matrix of the multi-view data in each view; For each view, adding a low-rank constraint to the data matrix to obtain a low-rank data matrix; For each low-rank data matrix corresponding to each view, the low-rank data matrix corresponding to each view is processed by a self-representation learning method to generate an affinity matrix of the multi-view data; Regularizing the affinity matrix of the multi-view data to obtain a processed affinity matrix; Performing independence processing on the processed affinity matrix by using the Hilbert-Schmidt independence criterion to obtain an independent affinity matrix; The independent affinity matrix is processed using an adaptive graph learning term to obtain a final data matrix corresponding to the multi-view data.
2. The method according to claim 1, characterized in that Before reconstructing the data in each view in different subspaces to obtain the data matrix of the multi-view data in each view, the method further includes: Data preprocessing is performed on the multi-view data to obtain processed multi-view data.
3. The method according to claim 2, characterized in that After preprocessing the multi-view data to obtain processed multi-view data, the method further includes: An original data matrix of the processed multi-view data in each view is generated.
4. The method according to claim 1, characterized in that: After the independent affinity matrix is processed by the adaptive graph learning item to obtain a final data matrix corresponding to the multi-view data, the method further includes: Based on the final data matrix, cluster labels of the multi-view data are generated.
5. A multi-view data clustering device, characterized in that: include: A reconstruction unit, configured to reconstruct the data in each view in different subspaces for the multi-view data, so as to obtain a data matrix of the multi-view data in each view; A constraint unit, used for adding a low-rank constraint to the data matrix for each view to obtain a low-rank data matrix; A first processing unit is configured to process the low-rank data matrix corresponding to each view by a self-representation learning method to generate an affinity matrix of the multi-view data; A second processing unit is used to perform regularization processing on the affinity matrix of the multi-view data to obtain a processed affinity matrix; A third processing unit is used to perform independence processing on the processed affinity matrix by using the Hilbert-Schmidt independence criterion to obtain an independent affinity matrix; The fourth processing unit is used to process the independent affinity matrix using the adaptive graph learning item to obtain a final data matrix corresponding to the multi-view data.
6. The device according to claim 5, characterized in that Also includes: The preprocessing unit is used to perform data preprocessing on the multi-view data to obtain processed multi-view data.
7. The device according to claim 6, characterized in that Also includes: The first generating unit is used to generate an original data matrix of the processed multi-view data in each view.
8. The device according to claim 5, characterized in that Also includes: The second generating unit is configured to generate cluster labels for the multi-view data based on the final data matrix.
9. An electronic device, characterized in that: include: one or more processors; a storage device having one or more programs stored thereon; When the one or more programs are executed by the one or more processors, the one or more processors are enabled to implement the method according to any one of claims 1 to 4.
10. A computer storage medium, characterized in that: A computer program is stored thereon, wherein when the computer program is executed by a processor, the method according to any one of claims 1 to 4 is implemented.