Large-scale multi-view tensor clustering method, device and apparatus based on linear kernel

Abstract

The present application relates to a large-scale multi-view tensor clustering method, device and equipment based on linear kernel. The method includes: obtaining multi-view data and a linear kernel matrix corresponding to each view data, performing kernel alignment on the linear kernel matrix to generate a corresponding potential data representation; stacking the potential data representation corresponding to each view data into a three-dimensional tensor, implementing a nuclear norm minimization constraint based on tensor singular value decomposition on the three-dimensional tensor to construct an objective function; solving the objective function using a pre-set linear optimization strategy to obtain the best potential data representation for each view data, splicing the best potential data representations of multi-view data to obtain a cluster representation, and clustering the cluster representations to obtain a clustering result of the multi-view data. The use of this method can improve the accuracy of multi-view clustering while maintaining an unsupervised learning framework.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a large-scale multi-view tensor clustering method, device and equipment based on linear kernel. Background Art

[0002] In recent years, a large number of multi-view clustering methods have used tensor technology to deal with high-order correlations between different data views, so they are also called multi-view tensor clustering. Initially, these methods calculated multiple subspace representation matrices from the data of each view and stacked them into a three-dimensional tensor. They specifically proposed a nuclear norm minimization constraint based on tensor singular value decomposition (t-SVD). Then, the tensor was constrained by a nuclear norm minimization constraint based on tensor singular value decomposition (t-SVD), thus proposing a multi-view subspace clustering method based on t-SVD. Experimental results show that this method significantly improves the clustering performance on public datasets. Based on this, subsequent studies have proposed a variety of constraints based on t-SVD, such as weighted tensor Schatten-p norm minimization constraints and enhanced tensor rank minimization constraints.

[0003] In addition, existing large-scale multi-view clustering methods process large-scale data by designing algorithm optimization strategies with linear complexity, which can be divided into two categories. The first method is based on matrix decomposition technology to decompose the data into coefficient matrix and basis matrix. The second method selects a small number of data samples, constructs a corresponding anchor graph for each view, and replaces the original data of the view for calculation, so as to reduce the computational complexity.

[0004] Although existing research has made significant progress, there are still two major problems that restrict the further improvement of multi-view clustering performance:

[0005] 1) The problem of illegal introduction of data labels. Existing multi-view tensor clustering methods all synchronously rotate tensors when using t-SVD-based constraints. In this process, the fast Fourier transform (FFT) introduces data sequence information, causing data labels to be mistakenly included in the clustering process, thus violating the basic principles of unsupervised learning.

[0006] 2) Incomplete integration of sample similarity. Existing large-scale multi-view clustering methods, although using matrix decomposition and anchor point technology, often only consider the similarity between some data samples and fail to fully integrate the similarity information of all samples, which limits the further improvement of clustering performance. Summary of the invention

[0007] Based on this, it is necessary to provide a large-scale multi-view tensor clustering method, device and equipment based on linear kernel to address the above technical problems.

[0008] A large-scale multi-view tensor clustering method based on linear kernel, the method comprising:

[0009] Acquire multiple view data and a linear kernel matrix corresponding to each view data, and perform kernel alignment on the linear kernel matrix to generate corresponding potential data representation;

[0010] stacking the potential data representation corresponding to each view data into a three-dimensional tensor, and applying a nuclear norm minimization constraint based on tensor singular value decomposition to the three-dimensional tensor to construct an objective function;

[0011] The objective function is solved by using a preset linear optimization strategy to obtain the best potential data representation of each view data, the best potential data representations of multiple view data are concatenated to obtain a cluster representation, and the cluster representations are clustered to obtain a clustering result of the multiple view data.

[0012] A large-scale multi-view tensor clustering device based on a linear kernel, the device comprising:

[0013] A parameter acquisition module, used to acquire multi-view data and a linear kernel matrix corresponding to each view data, and perform kernel alignment on the linear kernel matrix to generate a corresponding potential data representation;

[0014] A model building module, used for stacking the potential data representation corresponding to each view data into a three-dimensional tensor, and implementing a nuclear norm minimization constraint based on tensor singular value decomposition on the three-dimensional tensor to build an objective function;

[0015] The model solving module is used to solve the objective function by using a preset linear optimization strategy to obtain the best potential data representation of each view data, to concatenate the best potential data representations of multiple view data to obtain a cluster representation, and to cluster the cluster representations to obtain a clustering result of the multiple view data.

[0016] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0017] Acquire multiple view data and a linear kernel matrix corresponding to each view data, and perform kernel alignment on the linear kernel matrix to generate corresponding potential data representation;

[0018] stacking the potential data representation corresponding to each view data into a three-dimensional tensor, and applying a nuclear norm minimization constraint based on tensor singular value decomposition to the three-dimensional tensor to construct an objective function;

[0019] The objective function is solved by using a preset linear optimization strategy to obtain the best potential data representation of each view data, the best potential data representations of multiple view data are concatenated to obtain a cluster representation, and the cluster representations are clustered to obtain a clustering result of the multiple view data.

[0020] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the following steps:

[0021] Acquire multiple view data and a linear kernel matrix corresponding to each view data, and perform kernel alignment on the linear kernel matrix to generate corresponding potential data representation;

[0022] stacking the potential data representation corresponding to each view data into a three-dimensional tensor, and applying a nuclear norm minimization constraint based on tensor singular value decomposition to the three-dimensional tensor to construct an objective function;

[0023] The objective function is solved by using a preset linear optimization strategy to obtain the best potential data representation of each view data, the best potential data representations of multiple view data are concatenated to obtain a cluster representation, and the cluster representations are clustered to obtain a clustering result of the multiple view data.

[0024] The above-mentioned large-scale multi-view tensor clustering method, device and equipment based on linear kernel, by constructing a linear kernel matrix for each view data and performing kernel alignment, fully considers the similarity between all data samples, making the clustering result more accurate, stacking the potential data representation corresponding to each view data into a three-dimensional tensor, and implementing the objective function based on the nuclear norm minimization constraint of the tensor singular value decomposition to construct the three-dimensional tensor, which can avoid tensor rotation and ensure that the third dimension of the target tensor is equal to the number of views of the multi-view data, thereby eliminating the illegal introduction of data labels and ensuring the unsupervised nature of the clustering process. Finally, the present invention adopts an algorithm strategy based on linear optimization, with a linear computational complexity, which significantly improves the computational efficiency, is particularly suitable for large-scale data sets, and can obtain clustering results in a short time to meet real-time requirements. The embodiments of the present invention can improve clustering performance and computational efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 1 is a flow chart of a large-scale multi-view tensor clustering method based on linear kernel in one embodiment;

[0026] Figure 2 A schematic diagram of tensor rotation in one embodiment;

[0027] Figure 3302 is a structural block diagram of a large-scale multi-view tensor clustering device based on a linear kernel in one embodiment, wherein 302 represents a parameter acquisition module, 304 represents a model construction module, and 306 represents a model solving module;

[0028] Figure 4 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0030] In one embodiment, Figure 1 As shown, a large-scale multi-view tensor clustering method based on linear kernel is provided, comprising the following steps:

[0031] Step 102 : acquiring multi-view data and a linear kernel matrix corresponding to each view data, and performing kernel alignment on the linear kernel matrix to generate a corresponding potential data representation.

[0032] The data of each view can be represented by calculating its linear kernel matrix, which is a symmetric matrix used to measure the similarity between samples. The multi-view data is denoted as ( , , and Represents the number of views, samples and dimensions of view data), Recorded as The potential data representation corresponding to each view is ( ) is recorded as A three-dimensional tensor consisting of stacked potential representations, is the number of sample categories, The linear kernel of the view data can be expressed as

[0033] (1)

[0034] In order to extract consistent clustering structures from different views, the kernel matrix of each view needs to be adjusted so that they are aligned to a common representation space. This kernel alignment operation is achieved by introducing a latent data representation.

[0035] Step 104 , stacking the potential data representation corresponding to each view data into a three-dimensional tensor, and applying a nuclear norm minimization constraint based on tensor singular value decomposition to the three-dimensional tensor to construct an objective function.

[0036] t-SVD (Tensor Singular Value Decomposition) is a technique for decomposing tensors. It decomposes high-dimensional tensors into the product of multiple low-dimensional matrices, which can effectively reveal the underlying pattern or structure of the data. t-SVD can not only capture the low-order structural information in the tensor, but also maintain the correlation between different views.

[0037] like Figure 2 As shown in the figure, existing multi-view tensor clustering methods all synchronously rotate tensors when using t-SVD-based constraints. Since t-SVD requires the use of fast Fourier transform (FFT) on the third dimension of the target tensor, the data sequence information is integrated into the subsequent clustering process, thereby introducing data labels, which violates the basic settings of unsupervised learning.

[0038] It can be understood that the present invention does not perform tensor rotation to ensure that the third dimension of the data (i.e., the number of views of the multi-view data) is consistent with the dimension of the target tensor, thereby avoiding the illegal introduction of data labels. The method of the present invention does not introduce any data labels, but processes multi-view data by kernel alignment, mapping the data representation of each view to a common latent space. Through tensor singular value decomposition and nuclear norm minimization constraints, the focus is on the structural information of the data, rather than directly introducing external information through rotation or other means. In this way, the similarity relationship between multi-view data can be effectively retained, and the accuracy of clustering can be improved while maintaining the unsupervised learning framework.

[0039] Step 106, using a preset linear optimization strategy to solve the objective function, obtaining the best potential data representation for each view data, concatenating the best potential data representations for multiple view data to obtain a cluster representation, and clustering the cluster representations to obtain a clustering result for the multiple view data.

[0040] In the above-mentioned large-scale multi-view tensor clustering method based on linear kernel, by constructing a linear kernel matrix for each view data and performing kernel alignment, the similarity between all data samples is fully considered, so that the clustering result is more accurate, and the potential data representation corresponding to each view data is stacked into a three-dimensional tensor, and the three-dimensional tensor is implemented based on the tensor singular value decomposition The nuclear norm minimization constraint is constructed to construct the objective function, which can avoid tensor rotation and ensure that the third dimension of the target tensor is equal to the number of views of the multi-view data, thereby eliminating the illegal introduction of data labels and ensuring the unsupervised nature of the clustering process. Finally, the present invention adopts an algorithm strategy based on linear optimization, and the computational complexity is linear, which significantly improves the computational efficiency. It is particularly suitable for large-scale data sets and can obtain clustering results in a short time to meet real-time requirements. The embodiment of the present invention can improve clustering performance and computational efficiency.

[0041] In one embodiment, the objective function is:

[0042] (2)

[0043] in, For the View data, is the trace of the matrix, For the The potential data representation corresponding to the view data is is the linear kernel matrix, for dimensional identity matrix, Reason The potential data representations are stacked into a three-dimensional tensor with a dimension of , is the number of views, is the transpose operation, is a hyperparameter, is the nuclear norm based on the tensor singular value decomposition. In this embodiment, Need to be given before optimization.

[0044] In one embodiment, solving the objective function using a preset linear optimization strategy to obtain the best potential data representation for each view data includes: using auxiliary tensors, Lagrange multipliers and compensation parameters to decompose the nuclear norm in the objective function to obtain a new objective function; using a loop optimization method to update the potential data representation, auxiliary tensors and Lagrange multipliers in the new objective function until the preset iteration stop condition is met, stopping the iteration and outputting the best potential data representation for each view data; each update is achieved by performing singular value decomposition on the new objective function. In this embodiment, the objective function is solved by a loop optimization algorithm, and each round of optimization is divided into three parts: solving the potential data representation, solving the auxiliary tensors and updating the Lagrange multipliers.

[0045] Specifically, to solve formula (2), the method of the present invention proposes an optimization algorithm based on the augmented Lagrange multiplier method (ALM). The algorithm introduces an additional tensor As a new variable to assist optimization, formula (2) is rewritten as

[0046] (3)

[0047] Among them, the tensor is the Lagrange multiplier, is the compensation parameter.

[0048] In one embodiment, the use of a loop optimization method to update the potential data representation, auxiliary tensors and Lagrange multipliers in the new objective function includes: fixing the auxiliary tensors and Lagrange multipliers in the new objective function, and updating the potential data representation in the current new objective function by means of singular value decomposition to obtain a new potential data representation; fixing the potential data representation and Lagrange multipliers in the new objective function, and updating the auxiliary tensors in the current new objective function by means of singular value decomposition to obtain a new auxiliary tensor; updating the Lagrange multipliers in the new objective function according to the new potential data representation and the new auxiliary tensor to obtain a new Lagrange multiplier; if the iteration stop condition is not met, updating the compensation parameters and continuing to iterate the above steps.

[0049] In one embodiment, singular value decomposition is used to update the potential data representation in the current new objective function to obtain the new potential data representation, which includes: substituting fixed auxiliary tensors and Lagrange multipliers into the current new objective function to construct a first target matrix; the first target matrix includes current view data, potential data representation, and fixed auxiliary tensors and Lagrange multipliers; performing singular value decomposition on the first target matrix to obtain a first singular value decomposition result, and updating the potential data representation according to the first singular value decomposition result to obtain a new potential data representation.

[0050] 1) Specific, targeted Solving for variables, fixed tensors and Lagrange multipliers , formula (3) is equivalent to (4)

[0051] in, . Solve formula (4) using Algorithm 1:

[0052]

[0053] In one embodiment, the auxiliary tensor in the current new objective function is updated by singular value decomposition to obtain the new auxiliary tensor, which includes: substituting a fixed potential data representation and a Lagrange multiplier into the current new objective function to construct a second target matrix; the second target matrix includes compensation parameters, auxiliary tensors, and fixed potential data representations and Lagrange multipliers; performing singular value decomposition on the second target matrix to obtain a second singular value decomposition result, and updating the auxiliary tensor according to the second singular value decomposition result to obtain a new auxiliary tensor.

[0054] 2) Specific, targeted Solving for variables. Fixed latent data representation and Lagrange multipliers , formula (3) can be transformed into

[0055] (5)

[0056] in, The solution of formula (5) is

[0057] (6)

[0058] in, Represents a variable Perform an inverse Fourier transform along the third dimension, the variable No. forward slice for

[0059] (7)

[0060] in, , and for The product of the singular value decomposition of ,and .

[0061] 3) In one embodiment, for Solving for variables. Fixed latent data representation and tensors , the Lagrange multiplier It can be updated by:

[0062] (8)

[0063] In one embodiment, the iteration stop condition includes: calculating the function value of the new objective function based on the updated potential data representation, auxiliary tensor and Lagrange multiplier at the current iteration number, and calculating the objective function change value based on the function value at the current iteration number and the previous iteration number; when the objective function change value is less than or equal to a first threshold, and the norm value between the updated auxiliary tensor and the potential data representation at the current iteration number is less than or equal to a second threshold, the iteration stop condition is met.

[0064] Specifically, whether the convergence condition is met is determined by the following method:

[0065] and ;

[0066] in, For the The change value of the objective function at the iteration is, For the The change value of the objective function during iteration, is the first threshold, is the second threshold.

[0067] In a specific embodiment, the above-mentioned optimization algorithm based on the augmented Lagrange multiplier method (ALM) can be summarized as Algorithm 2 using pseudo code:

[0068]

[0069] The present invention is experimentally verified on the public datasets HW and BDGP. Among them, the comparison algorithms include RMKC (multi-view k-means clustering method), BMVC (Binary Multi-view Clustering), LMSC (Large-scale Multi-view Subspace Clustering), OPMC (One-pass Multi-view Clustering), EOMSC (Efficient One-pass Multi-view Subspace Clustering), ASR-ETR (anchor structure regularized multi-view subspace clustering by enhancing tensor rank minimization) and S2MVTC (scalable multi-view tensor clustering) algorithms in the literature. The large-scale multi-view tensor clustering method based on linear kernel is denoted as LMTC, and the clustering accuracy results obtained from the experiment are shown in Table 1.

[0070] Table 1 Clustering accuracy statistics of various methods

[0071]

[0072] It can be seen that the large-scale multi-view tensor clustering method based on linear kernel achieves the best clustering accuracy, which verifies the effectiveness of this method.

[0073] It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0074] In one embodiment, Figure 3 As shown, a large-scale multi-view tensor clustering device based on linear kernel is provided, comprising:

[0075] A parameter acquisition module 302 is used to acquire multi-view data and a linear kernel matrix corresponding to each view data, and perform kernel alignment on the linear kernel matrix to generate a corresponding potential data representation;

[0076] A model building module 304 is used to stack the potential data representation corresponding to each view data into a three-dimensional tensor, and implement a nuclear norm minimization constraint based on tensor singular value decomposition on the three-dimensional tensor to build an objective function;

[0077] The model solving module 306 is used to solve the objective function using a preset linear optimization strategy to obtain the best potential data representation of each view data, concatenate the best potential data representations of multiple view data to obtain a cluster representation, and cluster the cluster representations to obtain a clustering result of the multiple view data.

[0078] In one embodiment, the objective function is:

[0079] ;

[0080] in, For the View data, is the trace of the matrix, For the The potential data representation corresponding to the view data is is the linear kernel matrix, for dimensional identity matrix, Reason The potential data representations are stacked into a three-dimensional tensor with a dimension of , is the number of views, is the transpose operation, is a hyperparameter, is the nuclear norm based on the singular value decomposition of the tensor.

[0081] In one of the embodiments, the method is further used to decompose the nuclear norm in the objective function using auxiliary tensors, Lagrange multipliers and compensation parameters to obtain a new objective function; the potential data representation, auxiliary tensors and Lagrange multipliers in the new objective function are updated using a loop optimization method until a preset iteration stop condition is met, the iteration is stopped, and the optimal potential data representation of each view data is output; each update is achieved by performing singular value decomposition on the new objective function.

[0082] In one of the embodiments, it is also used to fix the auxiliary tensor and Lagrange multiplier in the new objective function, and use singular value decomposition to update the potential data representation in the current new objective function to obtain a new potential data representation; fix the potential data representation and Lagrange multiplier in the new objective function, and use singular value decomposition to update the auxiliary tensor in the current new objective function to obtain a new auxiliary tensor; update the Lagrange multiplier in the new objective function according to the new potential data representation and the new auxiliary tensor to obtain a new Lagrange multiplier; if the iteration stop condition is not met, update the compensation parameter and continue to iterate the above steps.

[0083] In one embodiment, the iteration stop condition includes: calculating the function value of the new objective function based on the updated potential data representation, auxiliary tensor and Lagrange multiplier at the current iteration number, and calculating the objective function change value based on the function value at the current iteration number and the previous iteration number; when the objective function change value is less than or equal to a first threshold, and the norm value between the updated auxiliary tensor and the potential data representation at the current iteration number is less than or equal to a second threshold, the iteration stop condition is met.

[0084] In one of the embodiments, it is also used to substitute fixed auxiliary tensors and Lagrange multipliers into the current new objective function to construct a first target matrix; the first target matrix includes current view data, potential data representation, and fixed auxiliary tensors and Lagrange multipliers; singular value decomposition is performed on the first target matrix to obtain a first singular value decomposition result, and the potential data representation is updated according to the first singular value decomposition result to obtain a new potential data representation.

[0085] In one of the embodiments, it is also used to substitute the fixed latent data representation and Lagrange multipliers into the current new objective function to construct a second target matrix; the second target matrix includes compensation parameters, auxiliary tensors, and fixed latent data representation and Lagrange multipliers; singular value decomposition is performed on the second target matrix to obtain a second singular value decomposition result, and the auxiliary tensor is updated according to the second singular value decomposition result to obtain a new auxiliary tensor.

[0086] For the specific definition of the large-scale multi-view tensor clustering device based on linear kernel, please refer to the definition of the large-scale multi-view tensor clustering method based on linear kernel above, which will not be repeated here. Each module in the above-mentioned large-scale multi-view tensor clustering device based on linear kernel can be implemented in whole or in part 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, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0087] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 4 As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a large-scale multi-view tensor clustering method based on a linear kernel is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a key, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0088] Those skilled in the art will understand that Figure 4 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0089] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method in the above embodiment when executing the computer program.

[0090] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method in the above embodiment are implemented.

[0091] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0092] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0093] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention. It should be pointed out that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A large-scale multi-view tensor clustering method based on linear kernel, characterized in that: The method comprises: Acquire multiple view data and a linear kernel matrix corresponding to each view data, and perform kernel alignment on the linear kernel matrix to generate corresponding potential data representation; stacking the potential data representation corresponding to each view data into a three-dimensional tensor, and applying a nuclear norm minimization constraint based on tensor singular value decomposition to the three-dimensional tensor to construct an objective function; The objective function is solved by using a preset linear optimization strategy to obtain an optimal potential data representation of each view data, the optimal potential data representations of multiple view data are concatenated to obtain a cluster representation, and the cluster representations are clustered to obtain a clustering result of the multiple view data; The objective function is: in, For the View data, is the trace of the matrix, For the The potential data representation corresponding to the view data is is the linear kernel matrix, for dimensional identity matrix, Reason The potential data representations are stacked into a three-dimensional tensor with a dimension of , is the number of views, is the transpose operation, is a hyperparameter, is the nuclear norm based on the singular value decomposition of the tensor; The adopting of a preset linear optimization strategy to solve the objective function to obtain the best potential data representation for each view data includes: The nuclear norm in the objective function is decomposed by using auxiliary tensors, Lagrange multipliers and compensation parameters to obtain a new objective function; A loop optimization method is used to update the potential data representation, auxiliary tensors and Lagrange multipliers in the new objective function until a preset iteration stop condition is met, then the iteration is stopped and the best potential data representation of each view data is output; each update is achieved by performing singular value decomposition on the new objective function.

2. The method according to claim 1, characterized in that Adopting a loop optimization method to update the potential data representation, auxiliary tensors and Lagrange multipliers in the new objective function includes: Fix the auxiliary tensor and Lagrange multiplier in the new objective function, and update the potential data representation in the current new objective function by using singular value decomposition to obtain a new potential data representation; Fixing the potential data representation and the Lagrange multiplier in the new objective function, and updating the auxiliary tensor in the current new objective function by means of singular value decomposition to obtain a new auxiliary tensor; updating the Lagrange multiplier in the new objective function according to the new potential data representation and the new auxiliary tensor to obtain a new Lagrange multiplier; If the iteration stop condition is not met, the compensation parameters are updated and the above steps are continued to be iterated.

3. The method according to claim 1, characterized in that The iteration stop condition includes: Calculate the function value of the new objective function according to the updated potential data representation, auxiliary tensor and Lagrange multiplier at the current iteration number, and calculate the change value of the objective function according to the function value at the current iteration number and the previous iteration number; When the objective function change value is less than or equal to the first threshold, and the norm value between the updated auxiliary tensor and the potential data representation at the current iteration number is less than or equal to the second threshold, the iteration stop condition is reached.

4. The method according to claim 2, characterized in that: The singular value decomposition method is used to update the potential data representation in the current new objective function, and the new potential data representation includes: Substituting the fixed auxiliary tensor and the Lagrange multiplier into the current new objective function to construct a first objective matrix; the first objective matrix includes the current view data, the potential data representation, the fixed auxiliary tensor and the Lagrange multiplier; Performing singular value decomposition on the first target matrix to obtain a first singular value decomposition result, and updating the potential data representation according to the first singular value decomposition result to obtain a new potential data representation.

5. The method according to claim 2, characterized in that: The auxiliary tensor in the current new objective function is updated by singular value decomposition, and the new auxiliary tensor includes: Substituting the fixed potential data representation and the Lagrange multiplier into the current new objective function to construct a second objective matrix; the second objective matrix includes compensation parameters, auxiliary tensors, and the fixed potential data representation and the Lagrange multiplier; Perform singular value decomposition on the second target matrix to obtain a second singular value decomposition result, and update the auxiliary tensor according to the second singular value decomposition result to obtain a new auxiliary tensor.

6. A large-scale multi-view tensor clustering device based on linear kernel, characterized in that: The device comprises: A parameter acquisition module, used to acquire multi-view data and a linear kernel matrix corresponding to each view data, and perform kernel alignment on the linear kernel matrix to generate a corresponding potential data representation; A model building module, used for stacking the potential data representation corresponding to each view data into a three-dimensional tensor, and implementing a nuclear norm minimization constraint based on tensor singular value decomposition on the three-dimensional tensor to build an objective function; A model solving module, used to solve the objective function by using a preset linear optimization strategy to obtain the best potential data representation of each view data, concatenate the best potential data representations of multiple view data to obtain a cluster representation, and cluster the cluster representations to obtain a clustering result of the multiple view data; The objective function is: in, For the View data, is the trace of the matrix, For the The potential data representation corresponding to the view data is is the linear kernel matrix, for dimensional identity matrix, Reason The potential data representations are stacked into a three-dimensional tensor with a dimension of , is the number of views, is the transpose operation, is a hyperparameter, is the nuclear norm based on the singular value decomposition of the tensor; The model solving module is also used to decompose the nuclear norm in the objective function using auxiliary tensors, Lagrange multipliers and compensation parameters to obtain a new objective function; use a loop optimization method to update the potential data representation, auxiliary tensors and Lagrange multipliers in the new objective function until a preset iteration stop condition is met, stop the iteration, and output the best potential data representation for each view data; each update is achieved by performing singular value decomposition on the new objective function.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5 are implemented.

Images